PDA

Просмотр полной версии : IvanHoe


Страницы : 1 2 3 4 5 6 7 8 9 10 11 12 [13] 14 15 16 17 18 19 20 21 22 23 24 25 26 27

DarkSide
21.07.2010, 18:32
Dear friends and compilers of Ivan engines.Many thanks for your efforts. I want to inform you that almost all Ivan compiles from version 57 to 53 (32 bits only-XP sp2, I don't know about 64 bits), lose on time in eng-eng matches in CB gui and so they are unusable.

I must desagree with you my friend, there are very stables versions, anyway..you can try with x+y time, and I´goona give u a example, IvanHoe-T Ghost2 is very know for his bad time managment, but if you test it on..for example 5'3" is among the best engines it beat Rybka 4 even in 64 bits and 8 cores!!!!:sm75:

Regards

Telepat
21.07.2010, 18:35
IvanHoe T55A{POPCNT} x64
IvanHoe T55A{SSE4.2} x64
hello! which version to use for the Core i7? POPCNT or SSE4.2 ?

DarkSide
21.07.2010, 19:17
Hi, seems that for beat Rybka is enough a "Old" Ivanhoe

System Windows XP SP 3, 32 bits
Intel Dual Core, 2.40 Hgz 2 GB Ram,
Gui: Arena 1.99 beta
King Gambit Set 120 Games
Control Time : 5'3"
No Ending Tablebases
Rybka 4 load with Lasker settings

IvanHoe-T Ghost2 vs Deep Rybka 4 +34-27=59

one assumes that T Ghost2 is mainly a analysis engine, but with incremental time
it plays very, very strong chess.:sm165:

Regards:D

Ipman
21.07.2010, 19:17
hello! which version to use for the Core i7? POPCNT or SSE4.2 ?

That's a good question Telepat!
Just run the internal benchmark.
How to do..dubble click the engine.exe file ,you get a dos window and type for example "go depth 20" the engine begins to calculate..you wait till it stops and check the nodes/sec.(nps) you get.
Do the same with the other engines and compare them..the highest nodes/sec. is normally the fastest one!

You can also run STS test to see which one is better !?

Ipman.

jaguar
21.07.2010, 19:29
That's a good question Telepat!
Just run the internal benchmark.
How to do..dubble click the engine.exe file ,you get a dos window and type for example "go depth 20" the engine begins to calculate..you wait till it stops and check the nodes/sec.(nps) you get.
Do the same with the other engines and compare them..the highest nodes/sec. is normally the fastest one!

You can also run STS test to see which one is better !?

Ipman.

Hi Ipman,
I want to do STS test in DF12 gui.
Can you please advice how to do.
Thanks.
Jag

Ipman
21.07.2010, 19:33
Hi Jag, you don't use Skype..then you have realtime help!

Behovodec
21.07.2010, 19:47
IvanHoe T55A
IvanHoe T55A{No SSE} w32
IvanHoe T55A{SSE3} w32
IvanHoe T55A{No SSE} x64
IvanHoe T55A{SSE3} x64
IvanHoe T55A{POPCNT} x64
IvanHoe T55A{SSE4.2} x64
Now, The Speed is very good
and other Small changes
http://www.speedyshare.com/files/234...anHoe_T55A.rar
Good luck

Mirrors:
http://www.mediafire.com/download.php?sm7tle0cwb2w7m3
http://www.4shared.com/file/j6L9WEgN/IvanHoe_T55A.html
http://rapidshare.com/files/408181737/IvanHoe_T55A.rar
http://ifile.it/b8r43qc
http://depositfiles.com/files/z68eqqz1a
http://www.multiupload.com/3QTE2D0USD

Vlad0
21.07.2010, 21:15
That's a good question Telepat!
Just run the internal benchmark.
How to do..dubble click the engine.exe file ,you get a dos window and type for example "go depth 20" the engine begins to calculate..you wait till it stops and check the nodes/sec.(nps) you get.
Do the same with the other engines and compare them..the highest nodes/sec. is normally the fastest one!

Ipman.

If You run this command twice (on the same engine) the results may differ
in range up to 10%! The same is true for "benchmark" command.
So this method can determine the quickest engine if their speeds differ
essentially. Which is not the case with Popcnt and SSE4.2 versions.
I suppose that Popcnt version is more universal -- it will run on both
Intel and AMD cpus. In case where CPU is i7 or i5 it is clear that
SSE4.2 version should be quicker. At least in theory. But there may be no
safely measurable difference in practice.

Rendruk
21.07.2010, 21:31
Random thought. . . why in the world do people still have 32-bit Operating systems? You can download 64 bit OS for free, and the technology is so old that you could probably pick up a dual core CPU for 30 dollars USD that would be an older model but still probably better than whatever the person with the 32 bit comp is running with. . .

jc.m
21.07.2010, 21:37
Random thought. . . why in the world do people still have 32-bit Operating systems?

In my case, this was because I got my computer 2 years ago, it was installed with Vista, and at my company, it was recommended to install 32 bits because of problems with drivers and software (I'm a dotNet developer).
Now, I'll probably install a Windows 7 64 bits, but I need to backup my data, and my hard-disk is full, so it's not so simple...

jc.m
21.07.2010, 22:18
Automated builds !

Here we go: I submitted my first automated builds.
There are 9 versions for 32 bits:
http://code.google.com/p/ivanhoebuilds/downloads/list

I also committed the sources, and you can grab them with SVN.
Instructions here:
http://code.google.com/p/ivanhoebuilds/wiki/HowToStart

What I have changed:
- the sources are exactly as the original Linux ones. Only one minor change: I removed INLINE in SMP_init.c, because this created a problem with Intel C++
- I added 2 solution files: one for Visual C++ and two for Intel C++ (click on the icproj for Intel)
- I created a Nant script, which creates all the builds automatically. Just install Nant from http://nant.sf.net (instructions in the SVN). The script is pretty advanced, since it detects the available compilers, and generates executables along with their PDB.

On my TODO list:
1) create 64 builds with Intel C++
2) add PGO builds
3) fix the 64 builds with VS, but it doesn't seem easy due to several problems (for example, _asm is refused with these compilers)

TCH
21.07.2010, 22:25
Random thought. . . why in the world do people still have 32-bit Operating systems? You can download 64 bit OS for free, and the technology is so old that you could probably pick up a dual core CPU for 30 dollars USD that would be an older model but still probably better than whatever the person with the 32 bit comp is running with. . .
For example, because the computer 90% of the time used for other tasks, including business applications. Time spent on reinstalling the system, as well as the risks associated with this, the licenses for installed software (not always license allows installation on another system), etc. It is not always possible replace the processor (eg, for laptop). We'll have to spend time on the drivers search and re-establishment of customary working environment, then we must put useful settings for a large number of programs for everyday using. Loss of working time will be calculated in days. And the advantage will be only 30% of the time spent on chess, ie several tens of minutes per day. Personally, I prefer to be reduced by 30% the number of games.

TCH
21.07.2010, 22:29
If You run this command twice (on the same engine) the results may differ in range up to 10%
I made this precedure many times and the repetability was much better then 10% if there is no problem with cash. But it's better to use "go movetime 60000" instead of "go depth 20" (may be only for my PC).

jc.m
21.07.2010, 22:33
I made this precedure many times and the repetability was much better then 10% if there is no problem with cash. But it's better to use "go movetime 60000" instead of "go depth 20" (may be only for my PC).

Also, I recommend that you maximize the DOS box before running the benchmarks, otherwise it's a benchmark of your graphic card, because it's very verbose.
Press left Alt and Return at the same time.

Oh, I just noticed that my computer refuses to toggle to full-screen with DOS !
I hate Vista !

Vlad0
21.07.2010, 22:49
I'm still struggling to make a minimal patch for compiling the Windows version and keeping the Linux version working.
I'm almost done, but right now, I'm testing new PopCnt functions, which are a little bit faster than the software one in Ivanhoe.


There are as well "hardwere" assembler code included in win32bit,h
and win64bit.h, which replaces the software emulation code if
one defines HAS_POPCN as compiler switch! And MS VC understands
this code for x32 vonfigurations. For x64 one is to #include <intrin.h>
and replace assembler code with "return __popcnt64(w)".


I'll probably finish a clean Windows build tomorrow, and at this moment, it would be great to include all your custom patches.

I've just checked out the code from Your repository to my local note.
There are yet no .vcxproj files in the sources. So one needs a lot of manual work yet to make source compilable by MS VC

PS Thank You once more for pointing to this great prog: TortoiseSVN.
It took me 2 days to learn to some extent how one is to use it. I created local
repository with 2 original "decembrist's" branches (v.53 and v.54). Now I'll
add 2 new branches (53bU and 54mU) with my modifications.
In fact, 53mU is ready now and compiles under Intel (though is yet unstable:
it falls asleep every 5-6 games and sleeps eternally after that).
And I intend to make 54bU by just merging "decembrist's" v.54 with "my"
53bU. Hoping that this merging will take much less time than "manual"
merging (file by file) which took me earlier 4--6 hours to do.

PPS After I send my post I saw, that You had updated Your repository.
S I'll repeat the chekout now to see, whether projects are in the source now.

jc.m
21.07.2010, 23:09
There are as well "hardwere" assembler code included in win32bit,h
and win64bit.h, which replaces the software emulation code if
one defines HAS_POPCN as compiler switch! And MS VC understands
this code for x32 vonfigurations. For x64 one is to #include <intrin.h>
and replace assembler code with "return __popcnt64(w)".


Yes, I'm well aware of that, and also that they are not very well optimized, since I implemented 2 new software POPCNT (one for SSE2 and one for SSE3), that I'll commit soon.

I was thinking about the _popcnt, which should work for 32 bits, and replace the ugly POPCNT(UINT64).


I've just checked out the code from Your repository to my local note.
There are yet no .vcxproj files in the sources. So one needs a lot of manual work yet to make source compilable by MS VC


No, just double-click on the .vcproj, and VS2010 will convert it to vcxproj (presumably, since I don't use VS2010).
I'm using VS2008 on my main computer.


In fact, 53mU is ready now and compiles under Intel (though is yet unstable:
it falls asleep every 5-6 games and sleeps eternally after that).
And I intend to make 54bU by just merging "decembrist's" v.54 with "my"
53bU. Hoping that this merging will take much less time than "manual"
merging (file by file) which took me earlier 4--6 hours to do.

Yes, and I recommend that you create branches to ease the merge.
I'll write some wiki pages to explain how it works.


Hint: if you grab the trunk, it will contain my version.
You can get the full history by grabbing the ivanhoe folder (just replace /trunk by /ivanhoe).

jc.m
21.07.2010, 23:24
Here are the 2 new POPCNT.

For SSE2:

#include <intrin.h>
#include <nmmintrin.h>
int POPCNTSSE2(UINT64 value) {
// Put masks in the main function to avoid reloading
const __m128i mask1 = _mm_set1_epi8(0x55);
const __m128i mask2 = _mm_set1_epi8(0x33);
const __m128i mask4 = _mm_set1_epi8(0x0F);

__m128i last = _mm_movpi64_epi64(value);
last = _mm_add_epi8( _mm_and_si128(last, mask1), _mm_and_si128(_mm_srli_epi16(last, 1), mask1) );
last = _mm_add_epi8( _mm_and_si128(last, mask2), _mm_and_si128(_mm_srli_epi16(last, 2), mask2) );
last = _mm_and_si128( _mm_add_epi8(last, _mm_srli_epi16(last, 4)), mask4 );
last = _mm_sad_epu8(last, _mm_setzero_si128());
return _mm_cvtsi128_si32(last);
}


For SSE3:

#include <intrin.h>
#include <nmmintrin.h>
int POPCNTSSE3(UINT64 value) {
// Put masks in the main function to avoid reloading
const __m128i mask_lo = _mm_set1_epi8(0x0F);
const __m128i mask_popcnt = _mm_setr_epi8(0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4);
__m128i last = _mm_movpi64_epi64(value);
__m128i lo = _mm_and_si128(last, mask_lo);
__m128i hi = _mm_and_si128(_mm_srli_epi16(last, 4), mask_lo);
lo = _mm_shuffle_epi8(mask_popcnt, lo);
hi = _mm_shuffle_epi8(mask_popcnt, hi);
last = _mm_add_epi8(lo, hi);
last = _mm_sad_epu8(last, _mm_setzero_si128());
return _mm_cvtsi128_si32(last);
}


In assembly language, the first routine takes 20 instructions, and the second one 14.
I have also the asm code for them, but they need optimization...

Right now, I'm a little bored, since I worked on the Nant script for almost 2 days...

Vlad0
21.07.2010, 23:31
No, just double-click on the .vcproj, and VS2010 will convert it to vcxproj (presumably, since I don't use VS2010).
I'm using VS2008 on my main computer.


I know how to convert .vcproj to .vcxproj. The problem is that
there is none of these files in grabbed code :(



Yes, and I recommend that you create branches to ease the merge.
I'll write some wiki pages to explain how it works.


Thank You in advance :)


Hint: if you grab the trunk, it will contain my version.
You can get the full history by grabbing the ivanhoe folder (just replace /trunk by /ivanhoe).

In fact, I've grabbed the full history folder.
Now I'll grab the other folder too (to a separate folder).

jc.m
21.07.2010, 23:37
I know how to convert .vcproj to .vcxproj. The problem is that
there is none of these files in grabbed code :(


This is because you didn't checkout the correct folder.

Hint: open Explorer, then right click on the folder and click on Properties.
Click on the tab "SubVersion"
It should say:
URL: http://ivanhoebuilds.googlecode.com/svn/trunk

If you don't want to install TortoiseSvn, you can install CollabNet, and use the svn.exe program to checkout the sources as:
svn checkout http://ivanhoebuilds.googlecode.com/svn/trunk/ ivanhoebuilds-read-only

Personally, I prefer AnkhSvn, since TortoiseSvn tends to slow down the computer (it tries hard to find svn folders when you use the explorer).

Sandi
21.07.2010, 23:45
Skulltrail Intel QX 9775 Yorkfield (8x4090 MHz)
Time 3'+0". Hash 128. GUI Fritz 12. Ponder off. Tablebases 5 pieces + Robbo TripleBases.book:HS-7moves.ctg.

IvanHoe T55A(SSE3)x64 - Deep Rybka 4 :
25 - 25 .50%-50%. (+7 -7 = 36)

The 50 games:

Vlad0
21.07.2010, 23:51
This is because you didn't checkout the correct folder.

Hint: open Explorer, then right click on the folder and click on Properties.
Click on the tab "SubVersion"
It should say:
URL: http://ivanhoebuilds.googlecode.com/svn/trunk


Thanks for this trick -- surely one of many I did not yet know.
I made it straightforwardly just creating a new empty folder and grabbing
the trunk into it.


If you don't want to install TortoiseSvn, you can install CollabNet, and use the svn.exe program to checkout the sources as:
svn checkout http://ivanhoebuilds.googlecode.com/svn/trunk/ ivanhoebuilds-read-only

Personally, I prefer AnkhSvn, since TortoiseSvn tends to slow down the computer (it tries hard to find svn folders when you use the explorer).

I've already installed Tortoise 2 days ago and am studing it since then.
Though I practically do not use Explorer (preferring Total commander)
and have not noted any slow down of the PC
I'll download both CollabNet & AnknSvn to see what they are.
Thank You!

jc.m
22.07.2010, 00:05
Thanks for this trick -- surely one of many I did not yet know.
I made it straightforwardly just creating a new empty folder and grabbing
the trunk into it.


Some other tricks:

Right-click on the folder, you have the following commands:

1) SVN Update (allows to grab the latest sources, and does a merge when possible)
2) SVN Commit (to send your sources). SVN does the commits by just committing patches (a few lines instead of the whole source)
3) TortoiseSVN.
Here are the functions in TortoiseSVN:
- Show Log: allows to display the full history of commits. You can do this on a single file to check what was changed and when
- Check for modifications: we do that before committing, to be sure that nobody changed our sources (giving conflicts)
- Branch: to create a branch
- Merge: to merge two branches
- Create patch: one of the most important functions ! it allows to create a small file containing all changes compared to the trunk (the trunk is the base code). Just send the patch, and anybody can merge it, by using apply patch
When you use BugZilla and fix a bug, you can attach a patch to the bug, it's in Bugzilla by default...
- another function when you right-click on a file: Blame, which gives who changed every line and when


I've already installed Tortoise 2 days ago and am studing it since then.
Though I practically do not use Explorer (preferring Total commander)
and have not noted any slow down of the PC
I'll download both CollabNet & AnknSvn to see what they are.
Thank You!

I'm using TC too ;-)
Collab is more for the svn.exe program (I don't use it for anything else).
Ankh is integrated to Visual Studio, so it's pretty useful when you use a local SVN server.
I can explain how to create a local server, if you are interested.

Rendruk
22.07.2010, 00:06
For example, because the computer 90% of the time used for other tasks, including business applications. Time spent on reinstalling the system, as well as the risks associated with this, the licenses for installed software (not always license allows installation on another system), etc. It is not always possible replace the processor (eg, for laptop). We'll have to spend time on the drivers search and re-establishment of customary working environment, then we must put useful settings for a large number of programs for everyday using. Loss of working time will be calculated in days. And the advantage will be only 30% of the time spent on chess, ie several tens of minutes per day. Personally, I prefer to be reduced by 30% the number of games.

I dunno what you mean by time and risk associated with installing a new OS. I've got it down to an art. I can get a new system up and running in 40 minutes without changes made post install, and 70 minutes after installing needed programs and making necessary changes (from installing things like fritz 12 to little things like changing my wallpaper), I even have a list of all of the windows settings, programs to install and registry tweaks I need to implement post install. I used to reinstall my system several times a week while working through customizing my own OS, nothing risky about it.

I dunno what programs you're referring to that have licences that are no longer valid after you reinstall windows because I've never had one of those programs and I illegally download all sorts of applications and games so I honestly have thousands of dollars worth of software (much more than the common person) and I've yet to run into that problem.

As for "searching for drivers", windows 7 comes with all the drivers you're going to need. No searching required except maybe getting the newest video card driver, which doesn't take very long.

Plus those 30% more games add up over the course of weeks, months, and years. An OS install might take you a day if you have no idea what you're doing (and if you don't that's your own fault, not the procedures), but it will eventually pay for itself in time gained.

The only thing you said that I really agreed with was that laptop users have a decent excuse. It can't be the money because I haven't held a job in years and I can afford a new processor every once in awhile with my college financial aid and the 50 bucks I might get at Christmas. . .

Sandi
22.07.2010, 00:15
Intel Xeon Octa (8x2660 MHz)
Time 3'+0". Hash 128. GUI Fritz 12. Ponder off.Tablebases 5 pieces + Robbo TripleBases. Book:HS-7moves.ctg.

IVANHOE T55A (NO SSE)x64 - DEEP RYBKA 4 : 30 - 20 !!!
60% -40% (+16 -6 =28) !!!!!!!!

THE ABSOLUTELY BEST RESULT OF A IVANHOE COMPILE AGAINST
DEEP RYBKA 4 TO 8 THREADS WITH 8 PHISICAL CORES.:sm165::sm165:

The 50 games:

Vlad0
22.07.2010, 00:33
Ankh is integrated to Visual Studio, so it's pretty useful when you use a local SVN server.
I can explain how to create a local server, if you are interested.

Oh, integration with VS is very important imo! So I'll install it surely.
I would be glad to read about creation of local SVN servers (or to have a link
to such info).

jc.m
22.07.2010, 00:55
Oh, integration with VS is very important imo! So I'll install it surely.
I would be glad to read about creation of local SVN servers (or to have a link
to such info).

I wanted to describe the process, but I realized that I forgot how to do it.
Here is how:
http://subversion.open.collab.net/articles/svnserve-service.htm
You need to install Collab.Net Subversion server, it's free.
Local SVN is useless in our case here.

I would prefer that we all use the SVN that I setup.
I'm waiting for the different compilers to test it, and once they understand clearly how it works, please send me your first patch.
We'll both validate the patch, and clarify the rules:

- the goal is to have always the original version from Linux
(try to not change the whole code, since it's painful to merge)
- you must use #defines for your patch, like

#define VLAD0

#ifdef VLAD0
Vlad0's code
#else
original code
#endif


Once this is well understood, I'll provide full rights to commit to the trunk (this means that you'll be able to change the source without me).
I'll do the merges in the first time, but I hope that Ippoliters will include our work.

slankamen
22.07.2010, 00:58
Q9550-XP64, Blitz:10' 0


1 Houdini 1.03a x64 4_CPU 3190 +12/-7/=81 52.50% 52.5/100
2 IvanHoe 9.53b x64 KLO 3185 +7/-12/=81 47.50% 47.5/100

Ahmed
22.07.2010, 01:25
Compare in the speed
The speed in T55A is nearly double than T55 in Fast PCs
so any of this engines is Number 1 and Number 2..........and Number 6 in the speed ?
IvanHoe T55A{No SSE} x64
IvanHoe T55A{SSE3} x64
IvanHoe T55A{POPCNT} x64
IvanHoe T55A{SSE4.2} x64
IvanHoe T55 x64
IvanHoe T55 i7

Thanks for Compare

checkmate4u
22.07.2010, 01:50
hi ahmed do u hv 32 bit for T55 A for cor 2 duo ??????

Ahmed
22.07.2010, 02:05
hi ahmed do u hv 32 bit for T55 A for cor 2 duo ??????
Look at this :
SSE4.2
Intel® Core™ i7 Processors
Intel® Core™ i5 Processors
Intel® Core™ i3 Processors
Intel® Xeon® 55XX series
Intel® Xeon® 56XX series
Intel® Xeon® 75XX series
SSE4.1
Intel® Xeon® 74XX series
Quad-Core Intel® Xeon 54XX, 33XX series
Dual-Core Intel® Xeon 52XX, 31XX series
Intel® Core™ 2 Extreme 9XXX series
Intel® Core™ 2 Quad 9XXX series
Intel® Core™ 2 Duo 8XXX series
Intel® Core™ 2 Duo E7200
SSE3
Quad-Core Intel® Xeon® 73XX, 53XX, 32XX series
Dual-Core Intel® Xeon® 72XX, 53XX, 51XX, 30XX series
Intel® Core™ 2 Extreme 7XXX, 6XXX series
Intel® Core™ 2 Quad 6XXX series
Intel® Core™ 2 Duo 7XXX (except E7200), 6XXX, 5XXX, 4XXX series
Intel® Core™ 2 Solo 2XXX series
Intel® Pentium® dual-core processor E2XXX, T23XX series
SSE3 Dual-Core Intel® Xeon® 70XX, 71XX, 50XX Series
Dual-Core Intel® Xeon® processor (ULV and LV) 1.66, 2.0, 2.16
Dual-Core Intel® Xeon® 2.8
Intel® Xeon® processors with SSE3 instruction set support
Intel® Core™ Duo
Intel® Core™ Solo
Intel® Pentium® dual-core processor T21XX, T20XX series
Intel® Pentium® processor Extreme Edition
Intel® Pentium® D
Intel® Pentium® 4 processors with SSE3 instruction set support
SSE2(default)
Intel® Xeon® processors
Intel® Pentium® 4 processors
Intel® Pentium® M
IA32
Intel® Pentium® III Processor
Intel® Pentium® II Processor
Intel® Pentium® Processor

kariconejito
22.07.2010, 02:08
donde estan los espejos para descargar ete super motor :sm78::sm78:

TCH
22.07.2010, 02:20
Rendruk, Not so simple. Install operating system and start playing chess I can for an hour or two. But to make dozens of programs, not just worked, but were comfortable and save time - quite another. I have considerable experience in this, but practice shows that the transition from "the system and the office is established, as well as all 20 programs that you use every day" to a situation in which "everything is conveniently configured and the computer became a docile instrument, tuned by specific user in a way that saves the most of his working time "is not a single day. Drivers - not everyone is in the distribution, especially for laptops, when you consider that I have connected about 15 external devices. Even just to organize the exchange of information on a local network with the distribution rights of tens of users and the organization of information security is spent not one hour.

TCH
22.07.2010, 03:23
Here we go: I submitted my first automated builds. There are 9 versions for 32 bits: http://code.google.com/p/ivanhoebuilds/downloads/list
Thank you for this build. I compared the speed of each compilation on my Core2Duo T9300 (SSE3 and 4.1 supported) and it was:

kN/s/core Program
686 Ivanhoe32_IA32_IC
739 Ivanhoe32_IA32_VC
697 Ivanhoe32_SSE2_IC
739 Ivanhoe32_SSE2_VC
685 Ivanhoe32_SSE3_IC
692 Ivanhoe32_SSE4.1_IC
681 Ivanhoe32_SSE_IC
745 Ivanhoe32_SSE_VC
683 Ivanhoe32_SSSE3_IC

The best result was shown by Ivanhoe32_SSE_VC, Ivanhoe32_SSE2_VC and (why??) Ivanhoe32_IA32_VC. So, for this CPU it will be better to use Ivanhoe32_SSE_VC (best speed) or some other compilation?

P.S. I made a small tournament with ultrashort TC 10 s/game (4 rounds, 32 games/engine, 360 games total)) and the winner was Ivanhoe32_SSSE3_IC (last position by speed!)

jc.m
22.07.2010, 03:50
The best result was shown by Ivanhoe32_SSE_VC, Ivanhoe32_SSE2_VC and (why??) Ivanhoe32_IA32_VC. So, for this CPU it will be better to use Ivanhoe32_SSE_VC (best speed) or some other compilation?

P.S. I made a small tournament with ultrashort TC 10 s/game (4 rounds, 32 games/engine, 360 games total)) and the winner was Ivanhoe32_SSSE3_IC (last position by speed!)

Yes, I noticed that the VC builds were pretty good on my computer too.
As Vlad0 explained, you cannot run a benchmark once and have a correct result. You need to run it a few times, and take the best time, though it's pretty difficult to determine.

SSSE3_IC means that Intel Compiler generates pretty good code for SSSE3 (it's not SSE3 !), but I'm not a specialist in tuning IC++.
I'm waiting for fine-tuning, if somebody knows it better than me.

Next steps are 64 bits build and PGO, although I don't know how I could optimize PGO automatically, and 64 bits compiles are problematic on a 32 bits computer...

BTW, I'm jealous, because SSE4.1 doesn't work on my computer.

TCH
22.07.2010, 04:33
As Vlad0 explained, you cannot run a benchmark once and have a correct result. You need to run it a few times, and take the best time, though it's pretty difficult to determine.
You can see two series of speed measurements, first number by "go movetime 30000", second by "go depth 20", both number kN/s for 1 core are in good relation and have only small difference:
681_687_Ivanhoe32_SSE_IC
683_685_Ivanhoe32_SSSE3_IC
685_689_Ivanhoe32_SSE3_IC
686_678_Ivanhoe32_IA32_IC
692_696_Ivanhoe32_SSE4.1_IC
697_672_Ivanhoe32_SSE2_IC
739_737_vanhoe32_IA32_VC
739_745_Ivanhoe32_SSE2_VC
745_748_Ivanhoe32_SSE_VC

kariconejito
22.07.2010, 04:45
You can see two series of speed measurements, first number by "go movetime 30000", second by "go depth 20", both number kN/s for 1 core are in good relation and have only small difference:
681_687_Ivanhoe32_SSE_IC
683_685_Ivanhoe32_SSSE3_IC
685_689_Ivanhoe32_SSE3_IC
686_678_Ivanhoe32_IA32_IC
692_696_Ivanhoe32_SSE4.1_IC
697_672_Ivanhoe32_SSE2_IC
739_737_vanhoe32_IA32_VC
739_745_Ivanhoe32_SSE2_VC
745_748_Ivanhoe32_SSE_VC

thc se que tu usas 32 bits dime cual es segun tu la mejor engine para una partida de 20 minutos

slankamen
22.07.2010, 05:07
Hola kariconejito,
en tu perfil personal dice que vives en Russia pero tus mensajes estan escritos en castellano.Tu eres de origen espanol o sole te gusta mucho el idioma.A mi me parece bien pero hay gente en este forum que no entiende espanol.
No es tan dificil usar google translator para escribir en ingles.
Saludos.

TCH
22.07.2010, 05:29
kariconejito, я не специалист в этом вопросе, поэтому не судите строго. На сегодняшний день в дебюте и миттельшпиле наилучшим на мой вкус является Houdini_w32_2CPU, ближе к эндшпилю, когда расчет на глубине 20-30 полуходов "дотягивается" до 6-фигурок при таком контроле я бы предпочел любую устойчивую версию Айвенго, работающую с трипл- и тоталбазами (например, IvanHoe T63D w32, IvanHoe 63Mod5Ub-PGO-x32, IvanHoe55mUI-x32, IvanHoe-B57d_14_intel_PGO и новее), или Рыбку 4 с 6-фигурками. Но пока Рыбкин расчет до Налимова не дотягивается - все же Айвенго, мне кажется, предпочтительнее благодаря быстрым триплбазам, размещенным в памяти. Если же всю партию надо играть одним движком - тогда Houdini. Когда к нему приделают роббобазы мало кто с ним сможет конкурировать.
Но я в этом не специалист, и мое мнение субъективно и непрофессионально. Лучше бы Вам проконсультироваться с более опытными игроками.
P.S. Извините, но я совершенно не знаю испанского.

TCH
22.07.2010, 05:53
Ahmed, when playing IvanHoe T55A{SSE3} w32 writes a log file UCIlog.txt (with or without GUI, and there is no option to switch of this logging. I think it's a bug, which can slow down the engine.

sirabc
22.07.2010, 06:49
Another way to test the speed of engines is to use Arena's automatic analysis. Just use the start position, add the engines you want tested. Create a separate log file to keep track of all tests.

kariconejito
22.07.2010, 07:25
kariconejito, я не специалист в этом вопросе, поэтому не судите строго. На сегодняшний день в дебюте и миттельшпиле наилучшим на мой вкус является Houdini_w32_2CPU, ближе к эндшпилю, когда расчет на глубине 20-30 полуходов "дотягивается" до 6-фигурок при таком контроле я бы предпочел любую устойчивую версию Айвенго, работающую с трипл- и тоталбазами (например, IvanHoe T63D w32, IvanHoe 63Mod5Ub-PGO-x32, IvanHoe55mUI-x32, IvanHoe-B57d_14_intel_PGO и новее), или Рыбку 4 с 6-фигурками. Но пока Рыбкин расчет до Налимова не дотягивается - все же Айвенго, мне кажется, предпочтительнее благодаря быстрым триплбазам, размещенным в памяти. Если же всю партию надо играть одним движком - тогда Houdini. Когда к нему приделают роббобазы мало кто с ним сможет конкурировать.
Но я в этом не специалист, и мое мнение субъективно и непрофессионально. Лучше бы Вам проконсультироваться с более опытными игроками.
P.S. Извините, но я совершенно не знаю испанского.

ok vale gracias por tu comentario xd
yo también tengo problema de idioma no pude configurar bien el registro por una cosa de idioma de echo me costo muchísimo suscribirme, quiero decirles que soy de Chile no de Rusia pero no se como cambiar el país y lo que leo es lo que traduce el google en todo caso la nacionalidad no me afecta para nada
bien amigos gracias por todo

rollno
22.07.2010, 08:09
Look at this :
SSE4.2
Intel® Core™ i7 Processors
Intel® Core™ i5 Processors
Intel® Core™ i3 Processors
Intel® Xeon® 55XX series
Intel® Xeon® 56XX series
Intel® Xeon® 75XX series
SSE4.1
Intel® Xeon® 74XX series
Quad-Core Intel® Xeon 54XX, 33XX series
Dual-Core Intel® Xeon 52XX, 31XX series
Intel® Core™ 2 Extreme 9XXX series
Intel® Core™ 2 Quad 9XXX series
Intel® Core™ 2 Duo 8XXX series
Intel® Core™ 2 Duo E7200
SSE3
Quad-Core Intel® Xeon® 73XX, 53XX, 32XX series
Dual-Core Intel® Xeon® 72XX, 53XX, 51XX, 30XX series
Intel® Core™ 2 Extreme 7XXX, 6XXX series
Intel® Core™ 2 Quad 6XXX series
Intel® Core™ 2 Duo 7XXX (except E7200), 6XXX, 5XXX, 4XXX series
Intel® Core™ 2 Solo 2XXX series
Intel® Pentium® dual-core processor E2XXX, T23XX series
SSE3 Dual-Core Intel® Xeon® 70XX, 71XX, 50XX Series
Dual-Core Intel® Xeon® processor (ULV and LV) 1.66, 2.0, 2.16
Dual-Core Intel® Xeon® 2.8
Intel® Xeon® processors with SSE3 instruction set support
Intel® Core™ Duo
Intel® Core™ Solo
Intel® Pentium® dual-core processor T21XX, T20XX series
Intel® Pentium® processor Extreme Edition
Intel® Pentium® D
Intel® Pentium® 4 processors with SSE3 instruction set support
SSE2(default)
Intel® Xeon® processors
Intel® Pentium® 4 processors
Intel® Pentium® M
IA32
Intel® Pentium® III Processor
Intel® Pentium® II Processor
Intel® Pentium® Processor

........................................ ....


Hi Ahmed , thanks for the information

I have a small doubt...

SSe3 is installed properly in my fritz11 ,

According to http://www.cpuid.com/softwares/cpu-z.html it shows

MMX,SSE(1,2,3,3S),EM64T

and i have Intel Celeron M 560 Processor which is not listed in ur list..

SO I am confused which engine to use...

any idea what is the difference between ....SSE3 and SSE3S ?

and what is EM64T?

carlospesce
22.07.2010, 08:11
Hi!
Are there posibilities for IvanHoev53 compiled by JR?
Thanks

Ahmed
22.07.2010, 11:26
Ahmed, when playing IvanHoe T55A{SSE3} w32 writes a log file UCIlog.txt (with or without GUI, and there is no option to switch of this logging. I think it's a bug, which can slow down the engine.
i run the engine on my system but i not found this file, are you use windows xp w32,anyway i think this is not bug, maybe i forget this when compiled this engine but this does not affect on the engine work
Hi Ahmed , thanks for the information

I have a small doubt...

SSe3 is installed properly in my fritz11 ,

According to http://www.cpuid.com/softwares/cpu-z.html it shows

MMX,SSE(1,2,3,3S),EM64T

and i have Intel Celeron M 560 Processor which is not listed in ur list..

SO I am confused which engine to use...

any idea what is the difference between ....SSE3 and SSE3S ?

and what is EM64T?
-with this Information {MMX,SSE(1,2,3,3S),EM64T} i think you can try IvanHoe T55A{SSE3}
-SSE3S is an incremental upgrade to SSE3, i think so
-EM64T is an abbreviation for Extended Memory 64 Technology and now known more commonly as Intel 64 or the x64

rollno
22.07.2010, 13:54
-with this Information {MMX,SSE(1,2,3,3S),EM64T} i think you can try IvanHoe T55A{SSE3}
-SSE3S is an incremental upgrade to SSE3, i think so
-EM64T is an abbreviation for Extended Memory 64 Technology and now known more commonly as Intel 64 or the x64

thanks buddy !

rybo
22.07.2010, 17:46
I must desagree with you my friend, there are very stables versions, anyway..you can try with x+y time, and I´goona give u a example, IvanHoe-T Ghost2 is very know for his bad time managment, but if you test it on..for example 5'3" is among the best engines it beat Rybka 4 even in 64 bits and 8 cores!!!!:sm75:

Regards

Yes my friend, I always play x'+y'' time and they lose "on time". Also, the Ghost series is version 63, which NEVER loses on time, in my system.

theo
22.07.2010, 19:25
Intel Xeon Octa (8x2660 MHz)
Time 3'+0". Hash 128. GUI Fritz 12. Ponder off.Tablebases 5 pieces + Robbo TripleBases. Book:HS-7moves.ctg.

IVANHOE T55A (NO SSE)x64 - DEEP RYBKA 4 : 30 - 20 !!!
60% -40% (+16 -6 =28) !!!!!!!!

THE ABSOLUTELY BEST RESULT OF A IVANHOE COMPILE AGAINST
DEEP RYBKA 4 TO 8 THREADS WITH 8 PHISICAL CORES.:sm165::sm165:

The 50 games:

None has rewarded this tests.I say "also" thanks" !:sm75:

WHMoweryJr
22.07.2010, 20:01
32-bit land, 2-core tests of UCI's
--------------------------------

Found some stuff in IvanHoe 9.53b code, and elsewhere.

Tested: {tourney stopped}
1) ExtraExtendInCheckPV - like EEIC in other versions,
but limited to in-PV king-moves only,
since the original EEIC was much weaker...
Still needs work:
Perhaps integrate in KingDanger AND
New KingAttack = IvanHoe KingAttack (combined_with) Stockfish KingAttack
This algorithm development topic can move to another thread,
but leaving in the UCI will allow EEIC development to continue...

2) ZugzwangDetect - implemented as a UCI bool - instead of PreprocessorDirective,
needs work...

3) MAGIC_BITBOARDS - implemented as a preprocessor directive,
modify Makefile
or command line with /D "MAGIC_BITBOARDS"
or in VC Project Properties - C/C++ - Preprocessor Definitions,
edit MAGIC_BITBOARDS in.
magic_mult.c needs to be included in project or command line or MakeFile.


Engine Score Iv Iv St Iv Iv S-B
1: IvanHoe-BetaWH_06 34.5/64 ················ 00==1==1=0=====0 1===10=====01=01 ===11=1===1=0==0 1==10======111== 1078.5
2: IvanHoe-BetaWH_06ZUG 32.5/64 11==0==0=1=====1 ················ ==10=00111===001 00=01========1== =1===1==0==0==== 1039.0
3: Stockfish-171-32-ja-fast 32.0/64 0===01=====10=10 ==01=11000===110 ················ ==0==1=10====11= 11100=00==1==100 1022.7
4: IvanHoe-BetaWH_06EEICPV 31.0/64 ===00=0===0=1==1 11=10========0== ==1==0=01====00= ················ 1====1=0===01=== 996.75
5: IvanHoe-BetaWH_06MBB 30.0/64 0==01======000== =0===0==1==1==== 00011=11==0==011 0====0=1===10=== ················ 971.50

160 of 16000 games played


Engine Score Iv Iv Iv St Iv S-B
1: IvanHoe-BetaWH_06 24.5/44 ··········· 00==1==1=0= ===11=1===1 1===10===== 1==10====== 522.75
2: IvanHoe-BetaWH_06ZUG 22.5/44 11==0==0=1= ··········· 00=01====== ==10=00111= =1===1==0== 492.75
3: IvanHoe-BetaWH_06EEICPV 21.5/44 ===00=0===0 11=10====== ··········· ==1==0=01== 1====1=0=== 470.50
4: Stockfish-171-32-ja-fast 21.0/44 0===01===== ==01=11000= ==0==1=10== ··········· 11100=00==1 466.00
5: IvanHoe-BetaWH_06MBB 20.5/44 0==01====== =0===0==1== 0====0=1=== 00011=11==0 ··········· 458.00

110 of 16000 games played


Engine Score Iv Iv Iv Iv St S-B
1: IvanHoe-BetaWH_06 22.0/40 ·········· 00==1==1=0 ===11=1=== 1==10===== 1===10==== 428.50
2: IvanHoe-BetaWH_06ZUG 20.5/40 11==0==0=1 ·········· 00=01===== =1===1==0= ==10=00111 407.25
3: IvanHoe-BetaWH_06EEICPV 20.0/40 ===00=0=== 11=10===== ·········· 1====1=0== ==1==0=01= 397.00
4: IvanHoe-BetaWH_06MBB 19.0/40 0==01===== =0===0==1= 0====0=1== ·········· 00011=11== 383.00
5: Stockfish-171-32-ja-fast 18.5/40 0===01==== ==01=11000 ==0==1=10= 11100=00== ·········· 376.75

100 of 16000 games played

Hash: 64M
Cores: 2
Level: Blitz 2/1
Hardware: Intel(R) Celeron(R) CPU E1200 @ 1.60GHz with 1,016 MB Memory 32-bit system
Operating system: Microsoft Windows Vista Home Edition Service Pack 1 (Build 6001)
PGN-File: C:\Program Files\Arena\Arena.pgn
E-Mail Address: WHMoweryJr@yahoo.com




Adding PANIC mode to timing
to see if that allows us
to run 2min 0sec games?

...making IvanHoe-BetaWH_07 soon...

Will post if timing is fixed,
apologies as i had to go back quite a few version numbers
to get to code without bug
and to add features/fixes/improvements
one at a time
adequately testing along the way.
For example, a MultiPv change needs a HaltSearch()
and that fixes the MultiPv analysis problem.
searchmoves needed a fix and now works, . . .

slankamen
22.07.2010, 22:39
Q9550-XP64, Blitz:10' 0


1 IvanHoe T55A.x64 3185 +10/-6/=64 52.50% 42.0/80
2 Houdini 1.03a x64 4_CPU 3190 +6/-10/=64 47.50% 38.0/80

Sandi
23.07.2010, 02:20
Skulltrail Intel QX 9775 Yorkfield (8x4090 MHz)
Time 3'+0". Hash 128. GUI Fritz 12. Ponder off. Tablebases 5 pieces + Robbo TripleBases.book:HS-7moves.ctg.

IvanHoe T55A(SSE3)x64 - Deep Rybka 4 :
25 - 25 .50%-50%. (+7 -7 = 36)

The 50 games:

Intel Xeon X5355 Octa (8X2660 MHz)
Time 3'+0".Hash 128.GUI Fritz 12.Ponder off.Tablebases 5 pieces + Robbo TripleBases. Book: IPPOLIT-7moves.ctg.

IvanHoe T55A(SSE3)x64 - Deep Rybka 4 : 11.5 -8.5 (+7 -4 =9)

The games:

kariconejito
23.07.2010, 03:24
¿alguien conoce esos motores tacticos ? yo tengo una version 3350b
parece que un nuevo oponente sale a la cancha :sm55::sm55::sm55:

jeandis
23.07.2010, 03:55
¿alguien conoce esos motores tacticos ? yo tengo una version 3350b
parece que un nuevo oponente sale a la cancha :sm55::sm55::sm55:

este motor no es muy bueno, tiene falla jugando las finales. Si lograrìa leer la tablas de Nalimov pudiera ser muy bueno pero no es así. Perdì elo jugando con el. A lo mejor hay un bug tanto en la version blitz que en la version Deep que le impide jugar bien las jugadas de finales. La presentación de los parametros es identico al de Houdini. Esta version x 64 bits no reconoce los 4 nucleos, solamente 2.

slankamen
23.07.2010, 04:43
Qi7- 960 XP.x64, Blitz:10' 0


1 IvanHoe T55A.x64SSE4.2 3195 +7/-2/=11 62.50% 12.5/20
2 Deep Rybka 4 SSE42 x64 3190 +2/-7/=11 37.50% 7.5/20

:sm15::sm36:

slankamen
23.07.2010, 05:19
AMD-X6.x64 Blitz:10', Blitz:10' 0


1 IvanHoe T55A.x64POPCNT 3195 +3/-2/=15 52.50% 10.5/20
2 Deep Rybka 4 x64 3185 +2/-3/=15 47.50% 9.5/20

Ahmed
23.07.2010, 18:17
http://s49.radikal.ru/i124/1007/2f/8465f3d9c4d9.bmp
429

Sandi
23.07.2010, 21:24
Skulltrail Intel QX 9775 Yorkfield (8X4090 MHz)
Time 5'+0".Hash 256. Ponder off. Tablebases 5 pieces.Book:IPOLLIT-7moves.ctg.

IvanHoe 54 JR - Deep Rybka 4 : 15-15. 50%-50% (+9-9=12)

Tomcass
23.07.2010, 21:44
2010-07IvanAhmed 2010

Quad 2.33
Gui: Fritz 12
Ponder: Off
No Triple or Robbo Bases
Book: HS Masterbook 2
Time Control: 3 min+ 1 sec.
4 cores

Deep Rybka 4 w32 - IvanHoe T55A 7.0 - 13.0 +0/=14/-6 35.00%

Sorry for posting this time only after 20 games, but this Ivan T55A seems really strong. Well done, Ahmed!!.

Regards,

Tom.

slankamen
23.07.2010, 23:14
Hi Tom,
in the Openchess in the engine testing section there is a thread on testing IvanHoe T55A.x64.
I think you should post this score there.
People from Rybka Lala land forum who by any chance go there should be able to see these scores because all the bad news is hidden from them.
Saludos.

checkmate4u
23.07.2010, 23:18
hello wanna know which ivan T55A u using i mean i hv to tested it but not found so impresive result
i hv tested ahmed T55A 32 bit for dual core processor in win 7 without books flitz 12
but did not find it so much good but ya i hv tested only small games......

WHMoweryJr
24.07.2010, 01:45
Before everyone goes nuts
because of this score,
it is possible in Ivan-land
to do something that excels,
and excels against some versions
or some engines,
but is not a real gain in ELO.

Complete testing is against
many many many
of the world's top engines...

It should be noted that this gain
is not manifested vs Stockfish 1.8,
so it probobly also
won't do much vs Rybka 4.

But here is the weird result:


Engine Score Iv Iv Iv St S-B
1: IvanHoe-BetaWH_07ManualBig 39.5/64 ···················· ·· ==0==10====1==0===0=1 10=====11==1==01===1= 1111110=1111110==1=110 1145.5
2: IvanHoe-BetaWH_07 32.0/64 ==1==01====0==1===1=0 ···················· ·· ==0=11===1===0=1=0=101 100=11==01==000=00110 1033.0
3: IvanHoe-BetaWH_07IvanhoeStyle 30.0/64 01=====00==0==10===0= ==1=00===0===1=0=1=010 ···················· ·· 1=0===0=011=====1=011 960.5
4: Stockfish-171-32-ja-fast 26.5/64 0000001=0000001==0=001 011=00==10==111=11001 0=1===1=100=====0=100 ···················· ·· 866.5

128 of 9600 games played
Name of the tournament: IvanHoe-BetaWH_07ManualBig
Site/ Country: EMACHINES-PC, United States
Level: Blitz 2/1
Hardware: Intel(R) Celeron(R) CPU E1200 @ 1.60GHz with 1,016 MB Memory
Operating system: Microsoft Windows Vista Home Edition Service Pack 1 (Build 6001)
PGN-File: C:\Program Files\Arena\Arena.pgn
Website:
E-Mail Address: WHMoweryJr@yahoo.com

Instead of: 100,300,310,500, 950
Used: 100,350,350,550,1050
For: P, N, B, R, Q

Yes, i often think outside the bun,
and find myself at Taco Bell,
eating crunchy tacos...


Добавлено через 9 минут
TIMING is improved:


Stockfish 1.8JA vs IvanHoe-BetaWH_07_w32 blitz2min0sec

1 Ivanhoe-BetaWH_07_w32 +54/-49/=56 51.57% 82.0/159
2 Stockfish 1.8 JA +49/-54/=56 48.43% 77.0/159

Timeouts:
Stockfish 1
Ivanhoe 0
51.27% for Ivan if timeout thrown out


Panic mode was hardwired ON
in this version,
starting at
less than
10 sec to go.

Tomcass
24.07.2010, 03:56
Hi Tom,
in the Openchess in the engine testing section there is a thread on testing IvanHoe T55A.x64.
I think you should post this score there.
People from Rybka Lala land forum who by any chance go there should be able to see these scores because all the bad news is hidden from them.
Saludos.

Done!

Saludos!

Tom.

halestorm
24.07.2010, 05:41
5+0
Arena, I7 oc'ed to 3.5 ghz with 6 GB RAM. Book on (7 moves). Bases ON. ExtraExtendInCheck=OFF; Try Large Pages=OFF
Win7-64


1: IvanHoe T55A{No SSE} x64 55.0/100 37-27-36 55%
2: Deep Rybka 4 SSE42 x64 45.0/100 27-37-36 45%


Strong win for Ivan.

Games attached.

TCH
24.07.2010, 06:01
5+0
Arena, I7 oc'ed to 3.5 ghz with 6 GB RAM. Book on (7 moves). Bases ON. ExtraExtendInCheck=OFF; Try Large Pages=OFF
Win7-64


1: IvanHoe T55A{No SSE} x64 55.0/100 37-27-36 55%
2: Deep Rybka 4 SSE42 x64 45.0/100 27-37-36 45%
Strong win for Ivan.

Games attached.
Bases ON for Ivanhoe, and what about Rybka?

jeandis
24.07.2010, 07:29
IvanHoe 9.53b x 64 is faster and better than T55A x64 noSSE version.

fustuk25
24.07.2010, 07:59
how do u check the speed of engine?and how do u know that is faster?do u have games between the 2 engines?

halestorm
24.07.2010, 10:37
Bases ON for Ivanhoe, and what about Rybka?

Of course. Rybka had access to 3-4-5 tablebases.

Ipman
24.07.2010, 12:11
Hi Bill, can we test your IvanHoe-BetaWH_07ManualBig compile and that someone can make a 64bit from it would be great!

JManion
24.07.2010, 12:44
OS: Windows xp 64
Quad 3.6ghz
Gui: DF 12
Ponder: Off
No Triple or Robbo Bases
Large Pages: Enabled
Book: perfect2009 10 moves
Time Control: 2 min+ 2 sec.

57 games

1 Houdini 1.03 x64 4_CPU +19/=33/-5 35.5/57
2 IvanHoe T55A x64 4_CPU +5/=33/-19 21.5/57

I just started this test and I will let it go overnight. It is not looking good for this version of IH.

Sandi
24.07.2010, 16:59
Intel Xeon X5355 Octa (8x2660 MHz)
Time 5'+0". Hash 256. Ponder off.Tablebases 5 pieces. GUI Fritz 12.
Book: IPPOLIT-7moves.ctg.

IvanHoe 57aUI-x64 - Deep Rybka 4 :
26.5 - 23.5. 53% - 47%. (+10 -7 =33).

WHMoweryJr
24.07.2010, 20:40
Name of the tournament: PAWN VALUE TEST

Engine Score
1: Stockfish-18-32-ja 56.5/152 37.17% ···················· ···················
2: IvanHoe-BetaWH_09PV100 25.5/38 67.11% 0====00111=101==11==11=1===11=1011111=
3: IvanHoe-BetaWH_09PV85 24.5/38 64.47% 0=11=1==0=01==11=101=1=1111=1===1=0101
4: IvanHoe-BetaWH_09PV95 23.0/38 60.53% =11===1=1=010=11=1011===111=10==0=0===
5: IvanHoe-BetaWH_09PV90 22.5/38 59.21% 010==1010==1=1=101=1==1==1=110011=====

152 of 6400 games played
Level: Blitz 2min/1sec
2-CORES: Intel(R) Celeron(R) CPU, E1200 @ 1.60GHz with 1,016 MB Memory
64M Hash
Titus.abk
OS: Microsoft Windows Vista Home Edition Service Pack 1 (Build 6001)
E-Mail Address: WHMoweryJr@yahoo.com


So, we see the baseline Pawn Value of 100 centi-pawns is probably best.
However not enough games were played to be sure,...

IvanHoe-BetaWH_09 baseline material:
P,N,B,R,Q == 100,315,325,515,980

==========================

Next test:

Pawn: 100
Knight: 300 + x
Bishop: 310 + x
Rook: 500 + x
Queen: 950 + x+x

Running x from 0 to 50,
prob 5 centi increments,
prob Stockfish 1.8 again...

checkmate4u
24.07.2010, 21:10
well run a small test for ivan vs houdini and rybka .........in dual core 32 bit win 7 flitz
12.no books ..4+2'........
houdini vs ry4....houdini played very nice good result with houdini .....
ivan vs ivan Ahmed T55a vs ivan 5.3b...i must say that ivan 5.3b is probably the best in ivan series so far.................
then ivan 5.3 b vs houdini ..............games is on but i must say there is nothing to choose between the two but if ivan is considered .. ivan 5.3b playes very nice probably the best and stongest engine on earth with houdini and rybka4 not so far behind him.........................
1. ivan 5.3b
2.houdini 1.3a
3.rybka4

halestorm
24.07.2010, 21:58
5+0 - 100 games
I7, Arena, 3.5 ghz, 64-bit windows7
All bases ON. Book ON. Same config as yesterday where Ivan won 55-45


Deep Rybka 4 SSE42 x64 54.5/100 39-30-31 55% +35
IvanHoe T55A{No SSE} x64 45.5/100 30-39-31 46% -28



Different from yesterday. Opposite. Why?

ExtraExtendInCheck=ON
Try Large Pages=ON

All other variables identical. Games enclosed.

Rendruk
24.07.2010, 22:39
This might seem weird to you, but twice now while testing Ivanhoe in STS the score has gone done by around 100 when I put 'try large pages' on.

It might be very bugged.

JManion
24.07.2010, 22:46
OS: Windows xp 64
Quad 3.6ghz
Gui: DF 12
Ponder: Off
No Triple or Robbo Bases
Large Pages: Enabled
Book: perfect2009 10 moves
Time Control: 2 min+ 2 sec.

86 games

1 Houdini 1.03 x64 4_CPU +32/=46/-8 55/86
2 IvanHoe T55A x64 4_CPU +8/=46/-32 31/86

I feel like something has to be wrong. I did not expect IH to preform this poorly. I was looking at some of the games, and most of the games it a very even position. Then IH makes a move or 2 that is not the best suddenly is down +1.3. From there IH just grinds it down without any problems. I assume Large pages is not working well in IH.

slankamen
24.07.2010, 23:35
Qi7- 960 XP.x64, Blitz:10' 0


1 Deep Rybka 4 SSE42 x64 3190 +4/-2/=14 55.00% 11.0/20
2 IvanHoe T55A.x64POPCNT 3195 +2/-4/=14 45.00% 9.0/20
:sm33:

Ipman
24.07.2010, 23:40
Thank you Slankamen..so it looks like the SSE4.2 version do it better on a i7 !

jc.m
25.07.2010, 00:20
1 Deep Rybka 4 SSE42 x64 3190 +4/-2/=14 55.00% 11.0/20
2 IvanHoe T55A.x64POPCNT 3195 +2/-4/=14 45.00% 9.0/20
:sm33:

I'm not surprised.
My tests show that Ivanhoe 999953 with 2 min +0, loses on time against Ivanhoe 999958m 95% of its games. In its last moves, instead of playing instantly, it takes its time to play, and I'm testing with the exact build from the sources, that is without any modification.

It's really difficult to understand what happens, since the code has no test (I'm used to work with TDD, aka Test Driven Development), and it's really hard to trace the problems, since the code is not always well designed (for example, the random routines are in array.c).
Running the engine and encountering these bugs just increase the feeling of insecurity and unstability.

I hope they'll add a function like "tests" à la "benchmark" to validate certain behaviors, like what to play when you have no time left, or check that all the underpromotions work effectively. What is missing is a way to reproduce the bugs, like forcing the random generator to a given value, so that we can observe what happens with a given random seed and a given position.

Just my 0.02 euros.

Anyway, I'm impressed by the courage of the compilers (I would like to call them modders, since they are changing the code more than compiling it).

Kudos !

slankamen
25.07.2010, 01:20
AMD Phenom II X6 1090T 32 bit, Blitz:10' 0


1 IvanHoe T55A.x32 3195 +3/-2/=15 52.50% 10.5/20
2 Deep Rybka 4 w32 3190 +2/-3/=15 47.50% 9.5/20

halestorm
25.07.2010, 04:14
2+2, 88 games, I7 64-bit Windows Arena

Bases ON for Ivan, books on
Large pages OFF for Houdini, ExtraExtendinCheck OFF


Houdini_x64_POPCNT_4CPU 65.0/88 52-10-26 74%
IvanHoe T55A{No SSE} x64 23.0/88 10-52-26 26%


Many time forfeits. Maybe my system? This Ivan did well against DR4-64

Games enclosed.

slankamen
25.07.2010, 04:55
Qi7- 960 XP.x64 4 GHZ 0


1 IvanHoe T55A.x64SSE4.2 3195 +1/-1/=18 50.00% 10.0/20 100.00
2 Deep Rybka 4 SSE42 x64 3190 +1/-1/=18 50.00% 10.0/20 100.00

slankamen
25.07.2010, 06:40
Q9550-XP64 3.5Ghz, Blitz:10' 0

1234567890123456
1 IvanHoe T54B.x64 3185 ½½½½½½½1½½½½½½½1 9.0/16
2 Deep Rybka 4 x64 3190 ½½½½½½½0½½½½½½½0 7.0/16

slankamen
25.07.2010, 06:55
AMD Phenom II X6 1090T 32 bit, Blitz:10' 0


1 Ivanhoe-BetaWH_09_w32 3195 +5/-3/=12 55.00% 11.0/20
2 Deep Rybka 4 w32 3190 +3/-5/=12 45.00% 9.0/20

slankamen
25.07.2010, 07:43
Today was not a good day for Ivanhoe.Tried different versions but the scores were below the expectation.

Qi7- 960 XP.x64 4 GHZ, Blitz:10' 0

1234567890
1 Deep Rybka 4 SSE42 x64 3190 ½½½½½½½1½½ 5.5/10
2 IvanHoe 9.53b-icc-001 x64 3195 ½½½½½½½0½½ 4.5/10

17
25.07.2010, 09:44
http://chess.cygnitec.com/engine/ivanhoe/

IvanHoe999953 -> updated [?] compilations 24.07.2010 г.

Ahmed
25.07.2010, 18:18
Today was not a good day for Ivanhoe.Tried different versions but the scores were below the expectation.
Without this would not be complete fun,this is the Life
Good luck to all

Ipman
25.07.2010, 18:45
Here some results from Ivanhoe 9.53b

Blitz 5m Core i7 920 @3.87Ghz , Blitz 5 0

IvanHoe 9.53b x64 2800 - Ivanhoe-B57dSSE4 2800 4.5 - 5.5 +1/-2/=7 45.00%
IvanHoe 9.53b x64 2800 - Houdini 1.03a x64 POPCNT 4_CP 2800 5.5 - 4.5 +1/-0/=9 55.00%
IvanHoe 9.53b x64 2800 - IvanHoe-T63Mini10 x64 2800 6.0 - 4.0 +3/-1/=6 60.00%
IvanHoe 9.53b x64 2800 - Houdini 1.01 x64 POPCNT 4_CPU 2800 6.0 - 4.0 +3/-1/=6 60.00%
IvanHoe 9.53b x64 2800 - IvanHoe-T63 Tactical x64 2800 5.5 - 4.5 +2/-1/=7 55.00%
IvanHoe 9.53b x64 2800 - Ivanhoe-B57d_whm01_w64 2800 5.0 - 5.0 +2/-2/=6 50.00%
IvanHoe 9.53b x64 2800 - IvanHoe-T63Mini9 x64 2800 5.0 - 5.0 +2/-2/=6 50.00%
IvanHoe 9.53b x64 2800 - FireBird 1.2 x64 2800 5.5 - 4.5 +2/-1/=7 55.00%
IvanHoe 9.53b x64 2800 - Houdini 1.02 x64 POPCNT 4_CPU 2800 5.0 - 5.0 +1/-1/=8 50.00%
IvanHoe 9.53b x64 2800 - FireBird 1.0 x64 2800 5.0 - 5.0 +0/-0/=10 50.00%
IvanHoe 9.53b x64 2800 - IvanHoe 57a4 2800 5.0 - 5.0 +0/-0/=10 50.00%
IvanHoe 9.53b x64 2800 - IvanHoe-B 63Mod12 2800 4.5 - 5.5 +1/-2/=7 45.00%
IvanHoe 9.53b x64 2800 - IvanHoe-B 63modB3s 2800 4.5 - 5.5 +0/-1/=9 45.00%
IvanHoe 9.53b x64 2800 - IvanHoe 9.57b x64 2800 5.0 - 5.0 +1/-1/=8 50.00%
IvanHoe 9.53b x64 2800 - IvanHoe T55 i7 2800 4.5 - 5.5 +0/-1/=9 45.00%
IvanHoe 9.53b x64 2800 - Fire 1.3 x64 2800 5.0 - 5.0 +0/-0/=10 50.00%
IvanHoe 9.53b x64 2800 - Stockfish 1.8 JA 64bit 2800 6.0 - 4.0 +2/-0/=8 60.00%
IvanHoe 9.53b x64 2800 - Igorrit 0.086v6 x64 2800 5.0 - 5.0 +1/-1/=8 50.00%
IvanHoe 9.53b x64 2800 - Stockfish 1.7 JA 64bit 2800 6.0 - 4.0 +3/-1/=6 60.00%
IvanHoe 9.53b x64 2800 - RobboLito 0.09 x64 2800 7.5 - 2.5 +5/-0/=5 75.00%
IvanHoe 9.53b x64 2800 - Stockfish 1.7.1 JA 64bit 2800 4.0 - 6.0 +1/-3/=6 40.00% -> with zugzwang detection On!!
IvanHoe 9.53b x64 2800 - IvanHoe-BetaWH_09_w64POPCNT 2800 4.0 - 6.0 +0/-2/=8 40.00%

Ivanhoe 9.53b takes the 3th place in my list after 4170 games!

Number one in this list is Houdini 1.03a x64 popcnt!!

Ipman.

slankamen
25.07.2010, 21:12
Intel(R) Core(TM)2 Quad CPU Q9550 @ 2.83GHz4x 3500 MHz with 4,096 MB Memory
Microsoft Windows XP 64 Bit Professional Service Pack 2 (Build 3790)
Fritz Benchmark:
Speed: 20.90
KNS: 10032
GUI: CB Rybka 3
Book: Perfect 2009 10 moves
Hash: 256
RB and TB: ON
Ponder: OFF

Blitz:10' 0


1 IvanHoe 9.53b x64 3185 +7/-5/=38 52.00% 26.0/50
2 Deep Rybka 4 x64 3190 +5/-7/=38 48.00% 24.0/50

This is new KLO version of IvanHoe T53.

WHMoweryJr
26.07.2010, 00:15
MATERIAL VALUES TESTING
==================

Parameter X was run from 0 to 50 centi-pawns.

Pawn: 100
kNight: 300 + X
Bishop: 310 + X
Rook: 500 + X
Queen: 950 + X + X


Engine Score
01: Stockfish-18-32-ja 73.5/180 ···············
02: IvanHoe-BetaWH_10X35 11.5/15 0111=1=11=1=11=
03: IvanHoe-BetaWH_10X25 10.5/15 0110=1110==1111
04: IvanHoe-BetaWH_10X30 9.5/15 0111=011100=11=
04: IvanHoe T57A w32 9.5/15 0==1==110101=11
04: IvanHoe-BetaWH_10X45 9.5/15 ==10=11=1=10==1
07: IvanHoe-BetaWH_10X00 8.5/15 =1=0=====0=111=
07: IvanHoe-BetaWH_10X40 8.5/15 ==000==1011111=
07: IvanHoe-BetaWH_10X15 8.5/15 =10==0101=110=1
10: IvanHoe-BetaWH_10X20 8.0/15 ==11=0=01=010=1
10: IvanHoe-BetaWH_10X05 8.0/15 0===01===001111
12: IvanHoe-BetaWH_10X10 7.5/15 ==0=10=1100==1=
13: IvanHoe-BetaWH_10X50 7.0/15 ==0==10=0=1==10

180 of 19200 games played
2-cores
Titus.abk
64M hash
Level: Blitz 2/1
Hardware: Intel(R) Celeron(R) CPU E1200 @ 1.60GHz with 1,016 MB Memory
Operating system: Microsoft Windows Vista Home Edition
E-Mail Address: WHMoweryJr@yahoo.com


Engine Score

07: IvanHoe-BetaWH_10X00 8.5/15
10: IvanHoe-BetaWH_10X05 8.0/15
12: IvanHoe-BetaWH_10X10 7.5/15
07: IvanHoe-BetaWH_10X15 8.5/15
10: IvanHoe-BetaWH_10X20 8.0/15
03: IvanHoe-BetaWH_10X25 10.5/15
04: IvanHoe-BetaWH_10X30 9.5/15
02: IvanHoe-BetaWH_10X35 11.5/15
07: IvanHoe-BetaWH_10X40 8.5/15
04: IvanHoe-BetaWH_10X45 9.5/15
13: IvanHoe-BetaWH_10X50 7.0/15
04: IvanHoe T57A w32 9.5/15



At this point,
X seems to want to be 35 centi-pawns,
but will continue for awhile so,
second half results can confirm first half
ELSE
continue {unless Stockfish learning[?] is tampering...}.

Sandi
26.07.2010, 00:38
Skulltrail Intel QX 9775 Yorkfield (8X4090 MHz).
Time 3'+0". Hash 128. Ponder off. GUI Fritz 12. Tablebases 5 pieces + RobboTripleBases. book:HS-7moves.ctg.

IvanHoe 9.53b x64 - Deep Rybka 4 :
25- 25 . 50%-50%. (+8 -8 =34)

The 50 games:

Tomcass
26.07.2010, 01:09
So is there a doubt that Houdini 1.03a is the strongest engine in the world now?

In my opinion, your statement is true if you add: 'at time controls no longer than 5 minutes'. At T.C. longer than 15 minutes I think the strongest engine in the world now is Deep Rybka 4. We need some additional work to get an engine able to overcome DR4 at longer time controls.

Regards,

Tom.

halestorm
26.07.2010, 01:17
20+0 per game, each side
I7-920 at 3.5 ghz, 6 gb ram
Bases ON, Book ON


IvanHoe T55A{No SSE} x64 - Deep Rybka 4 SSE42 x64 : 11.5/23 5-5-13


EVEN in long TCs

halestorm
26.07.2010, 01:59
We need some additional work to get an engine able to overcome DR4 at longer time controls.


Where does "longer time controls" begin? 10 min games? 20 min games? 40+4?

Tomcass
26.07.2010, 02:23
Where does "longer time controls" begin? 10 min games? 20 min games? 40+4?

Until today I have not found any engine able to consistently beat DR4 at TC of 10' 0 . Anything above this T.C. has to been proven yet. Not to say anything at a normal tournement game TC.

Tom.

... IMHO consistently means 'with a statistically significant advantage in a test of not least than 200 games'.

slankamen
26.07.2010, 02:28
Hardware: Intel(R) Core(TM) i7 CPU 960 @ 3.20GHz8x 4005 MHz with 4,096 MB Memory
Operating system: Microsoft Windows XP 64 Bit Professional Service Pack 2 (Build 3790)
Fritz Benchmark:
Speed: 24.98
KNS: 11988
GUI: CB Rybka 3
Book: Perfect 2009 10 moves
Hash: 256
RB and TB: ON
Ponder: OFF

Blitz:10' 0


1 Deep Rybka 4 SSE42 x64 3190 +4/-2/=14 55.00% 11.0/20
2 Ivanhoe-BetaWH_09_w64 3195 +2/-4/=14 45.00% 9.0/20

BTO7
26.07.2010, 02:34
MATERIAL VALUES TESTING
==================

Parameter X was run from 0 to 50 centi-pawns.

Pawn: 100
kNight: 300 + X
Bishop: 310 + X
Rook: 500 + X
Queen: 950 + X + X


Engine Score
01: Stockfish-18-32-ja 73.5/180 ···············
02: IvanHoe-BetaWH_10X35 11.5/15 0111=1=11=1=11=
03: IvanHoe-BetaWH_10X25 10.5/15 0110=1110==1111
04: IvanHoe-BetaWH_10X30 9.5/15 0111=011100=11=
04: IvanHoe T57A w32 9.5/15 0==1==110101=11
04: IvanHoe-BetaWH_10X45 9.5/15 ==10=11=1=10==1
07: IvanHoe-BetaWH_10X00 8.5/15 =1=0=====0=111=
07: IvanHoe-BetaWH_10X40 8.5/15 ==000==1011111=
07: IvanHoe-BetaWH_10X15 8.5/15 =10==0101=110=1
10: IvanHoe-BetaWH_10X20 8.0/15 ==11=0=01=010=1
10: IvanHoe-BetaWH_10X05 8.0/15 0===01===001111
12: IvanHoe-BetaWH_10X10 7.5/15 ==0=10=1100==1=
13: IvanHoe-BetaWH_10X50 7.0/15 ==0==10=0=1==10

180 of 19200 games played
2-cores
Titus.abk
64M hash
Level: Blitz 2/1
Hardware: Intel(R) Celeron(R) CPU E1200 @ 1.60GHz with 1,016 MB Memory
Operating system: Microsoft Windows Vista Home Edition
E-Mail Address: WHMoweryJr@yahoo.com


Engine Score

07: IvanHoe-BetaWH_10X00 8.5/15
10: IvanHoe-BetaWH_10X05 8.0/15
12: IvanHoe-BetaWH_10X10 7.5/15
07: IvanHoe-BetaWH_10X15 8.5/15
10: IvanHoe-BetaWH_10X20 8.0/15
03: IvanHoe-BetaWH_10X25 10.5/15
04: IvanHoe-BetaWH_10X30 9.5/15
02: IvanHoe-BetaWH_10X35 11.5/15
07: IvanHoe-BetaWH_10X40 8.5/15
04: IvanHoe-BetaWH_10X45 9.5/15
13: IvanHoe-BetaWH_10X50 7.0/15
04: IvanHoe T57A w32 9.5/15



At this point,
X seems to want to be 35 centi-pawns,
but will continue for awhile so,
second half results can confirm first half
ELSE
continue {unless Stockfish learning[?] is tampering...}.

So if im reading this right the most efficient material values from stock settings would be K=335 B=345 R=535 Q=1020 and of course assuming .35 holds? Interesting tests there you doing to say the least ....thanks

Regards
BT

slankamen
26.07.2010, 05:44
AMD Phenom(tm) II X6 1090T Processor-6x 3.2 GHz 3,328 MB Memory
Microsoft Windows XP Professional Service Pack 3 (Build 2600)
Fritz Benchmark:
Speed: 23.35
KNS: 11207
GUI: CB Rybka 3
Book: Perfect 2009 10 moves
Hash: 256
RB and TB: ON
Ponder: OFF

Blitz:10' 0


1 IvanHoe T55A.x32 3195 +9/-5/=36 54.00% 27.0/50
2 Deep Rybka 4 w32 3190 +5/-9/=36 46.00% 23.0/50

slankamen
26.07.2010, 06:34
AMD Phenom(tm) II X6 1090T Processor6x 3895 MHz with 4,096 MB Memory
Microsoft Windows XP 64 Bit Professional Service Pack 2 (Build 3790)
Fritz Benchmark:
Speed: 28.33
KNS: 13599
GUI: CB Rybka 3
Book: Perfect 2009 10 moves
Hash: 256
RB and TB: ON
Ponder: OFF
Blitz:10' 0


1 IvanHoe 9.53b x64 KLO 3195 +11/-10/=29 51.00% 25.5/50
2 Deep Rybka 4 x64 3185 +10/-11/=29 49.00% 24.5/50

WHMoweryJr
26.07.2010, 06:52
MATERIAL VALUES TESTING

P = 100
N = 300 + X
B= 310 + X
R= 500 + X
Q = 950 + X + X


Engine Score St
01: Stockfish-18-32-ja 103.0/277 ···················· ····
02: IvanHoe-BetaWH_10X35 17.0/23 0111=1=11=1=11=110=11==
03: IvanHoe-BetaWH_10X30 15.5/23 0111=011100=11=1=11=011
03: IvanHoe-BetaWH_10X40 15.5/23 ==000==1011111=11111110
05: IvanHoe-BetaWH_10X45 15.0/23 ==10=11=1=10==111=0==11
05: IvanHoe T57A w32 15.0/23 0==1==110101=111111=100
07: IvanHoe-BetaWH_10X00 14.5/23 =1=0=====0=111==1==11=1
07: IvanHoe-BetaWH_10X50 14.5/23 ==0==10=0=1==101111=111
09: IvanHoe-BetaWH_10X25 14.0/23 0110=1110==11111=0100==
10: IvanHoe-BetaWH_10X05 13.5/23 0===01===00111110111=10
10: IvanHoe-BetaWH_10X15 13.5/23 =10==0101=110=1111001==
12: IvanHoe-BetaWH_10X10 12.5/23 ==0=10=1100==1==11=0011
12: IvanHoe-BetaWH_10X20 12.5/23 ==11=0=01=010=111===0==

276 of 19200 games played
Site/ Country: EMACHINES-PC, United States
Level: Blitz 2'/1"
64M
Titus.abk
Hardware: Intel(R) Celeron(R) CPU 2-cores @1.60GHz, 1,016 MB Memory
Operating system: Microsoft Windows Vista Home Edition Service Pack 1
E-Mail Address: WHMoweryJr@yahoo.com



So at X = 35 we get:
P = 100
N = 335
B= 345
R= 535
Q = 1020

Bringing the next test:

P = 100
N = 335
B= 345
R= 535 + Y
Q = 1020 + Y + Y

Y running from -25 to 25, every 5 centi's

slankamen
26.07.2010, 08:05
Intel(R) Core(TM) i7 CPU 960 @ 3.20GHz8x 4.005 GHz 4,096 MB Memory
Windows 7 Professional Professional (Build 7600)
Fritz Benchmarks:
Speed: 24.89
KNS: 11946
GUI: CB Rybka 3
Hash: 256
Book: Perfect 2009 10 moves
RB and TB: ON
Ponder: OFF
Blitz:10' 0


1 IvanHoe T55A.x64SSE4.2 3190 +6/-0/=14 65.00% 13.0/20
2 Houdini 1.01 x64 4_CPU 2890 +0/-6/=14 35.00% 7.0/20

halestorm
26.07.2010, 08:49
STS12
5 sec per
12 tests of 100 each = total 1200 points avail
Tested on I7 with POPCNT


Engine 1 2 3 4 5 6 7 8 9 10 11 12 TOTALS
IvanHoe T55A{No SSE} x64 95 90 90 90 87 91 84 83 83 96 90 90 1069
IvanHoe T55A{POPCNT} x64 96 90 91 87 86 89 85 79 85 95 93 90 1066
Deep Rybka 4 SSE42 x64 95 91 92 90 94 87 87 79 79 89 88 86 1057

Sandi
26.07.2010, 09:00
Long time control with different books.
~~~~~~~~~~~~~~~~~~~~~~~~
Intel Xeon x5355 Octa (8x2660 MHz).
Time 30'+0".
Hash 1024.Ponder off. GUI Fritz 12.Nalimov Tablebases 3-4-5 + RobboTriple Bases.
Books: IPPOLIT-7moves.ctg for IvanHoe.
" Rybka 4.ctg (Jiri Dufek)up to 7 moves for Deep Rybka.

IvanHoe T55A (no SSE) - Deep Rybka 4.
11 - 9 ( + 5 - 3 = 12)

sirabc
26.07.2010, 21:19
From http://www.open-chess.org/viewtopic.php?f=7&t=470&start=10#p4443

kingliveson » Sun Jul 25, 2010 6:55 am
IvanHoe 999953 beta has been re-uploaded yet again. The files has been reloaded about 5 times since initial upload and last upload was about 15 minutes ago. If it is yet re-uploaded again, it will probably be the last one.

http://chess.cygnitec.com/engine/ivanhoe/

WHMoweryJr
26.07.2010, 21:23
More IvanHoe Differences Testing

Minor differences between IvanHoe 53, 55 & 57 are being investigated...


Engine Score
1: Stockfish-18-32-ja 43.5/108 40.28% ··················
2: IvanHoe-BetaWH_11ZUG 14.0/18 77.78% 01111==1=111===111
3: IvanHoe-BetaWH_11L5S 12.0/18 66.67% 11=1011===0==10111
4: IvanHoe-BetaWH_11EPE 10.5/18 58.33% =11=01=0==1101=1=0
5: IvanHoe-BetaWH_11 10.0/18 55.56% 0=11=1===0==01===1
5: IvanHoe-BetaWH_11ETE 10.0/18 55.56% =0=1=110==10=0=11=
7: IvanHoe-BetaWH_11EEIC 8.0/18 44.44% 011=0110====0===00

Name of the tournament: IvanHoe-BetaWH_11 UCI's
E-Mail Address: WHMoweryJr@yahoo.com

ZUG: ZugzwangDetect
L5S: Low999955mSkip
EPE: ExtraPasserExtensions
ETE: ExtraTacticlExtensions
EEIC: ExtraExtendInCheckPV

Conclusions:
Hardwire ZUG ON.
Remove UCI ExtraExtendInCheckPV & remove EXTEND_IN_CHECK.



ZugzwangDetect is from 999953 and tests strong, itsa keeper.

Low999955mSkip is from 999955 and will remain a UCI
to see how it does when zug detect is always on. It prohibits
the skipping of King-moves in low_depth.c
More testing is needed...

ExtraPasserExtensions gives us improvement vs R4
but Stockfish 1.8 seems not to notice it...
More testing needed...
It is from DamirsRybkaKiller

ExtraTacticalExtensions gives us improvement vs R4
but Stockfish 1.8 seems not to notice it...
More testing needed...
ETE is from 999955.

checkmate4u
26.07.2010, 21:27
why uploading it again and again is there any changes in it or tis just the same as the last one????????????

thanks

slankamen
26.07.2010, 21:39
This was one of the latest version but did not do to well.
Things can be tricky with compiling.

Q9550-XP64 3.5Ghz, Blitz:10' 0


1 Deep Rybka 4 x64 3190 +21/-10/=39 57.86% 40.5/70
2 IvanHoe 9.53b x64 KLO R 3185 +10/-21/=39 42.14% 29.5/70

TCH
26.07.2010, 21:51
why uploading it again and again is there any changes in it or tis just the same as the last one????????????

thanks
It's not the same file, there are some changes.

slankamen
26.07.2010, 22:42
Intel(R) Core(TM) i7 CPU 960 @ 3.20GHz8x 4.005 GHz 4,096 MB Memory
Windows 7 Professional Professional (Build 7600)
Fritz Benchmarks:
Speed: 24.89
KNS: 11946
GUI: CB Rybka 3
Hash: 256
Book: Perfect 2009 10 moves
RB and TB: ON
Ponder: OFF
Blitz:10' 0


1 Houdini 1.03a x64 POPCNT 4_CP 3190 +3/-0/=17 57.50% 11.5/20
2 IvanHoe T55A.x64SSE4.2 3190 +0/-3/=17 42.50% 8.5/20

WHMoweryJr
27.07.2010, 01:36
why uploading it again and again is there any changes in it or tis just the same as the last one????????????

thanks

Code looks totally un-changed.

Builds are newer,
32-bit build is about 1.5% faster than the older 32-bit build...

Porque?
Porque nps rules...

Ahmed
27.07.2010, 02:03
http://i047.radikal.ru/1007/de/186a30f11f11.bmp
444

IvanHoe 53 seems good, so i will make update for IvanHoe T53 with the same speed in IvanHoe T55A to see what will happen

slankamen
27.07.2010, 03:22
Intel(R) Core(TM) i7 CPU 960 @ 3.20GHz8x 4.005 GHz 4,096 MB Memory
Windows 7 Professional Professional (Build 7600)
Fritz Benchmarks:
Speed: 24.89
KNS: 11946
GUI: CB Rybka 3
Hash: 256
Book: Perfect 2009 10 moves
RB and TB: ON
Ponder: OFF


Blitz:10' 0


1 Houdini 1.03a x64 POPCNT 4_CP 3190 +3/-0/=17 57.50% 11.5/20
2 Ivanhoe-BetaWH_12_w64POPCNT 3190 +0/-3/=17 42.50% 8.5/20

This was experimental version but so far Ivanhoe is having problems beating Houdini 1.03a

TCH
27.07.2010, 05:21
IvanHoe 53 seems good, so i will make update for IvanHoe T53 with the same speed in IvanHoe T55A to see what will happen
Ahmed, add autoloading of Robbobases to your new compilation, please!

yayo
27.07.2010, 05:41
http://i047.radikal.ru/1007/de/186a30f11f11.bmp
444

IvanHoe 53 seems good, so i will make update for IvanHoe T53 with the same speed in IvanHoe T55A to see what will happen

Ahmed what is this Ivanhoe 9.53b x64 and where can i find it?

onedrey
27.07.2010, 05:48
Ahmed, add autoloading of Robbobases to your new compilation, please!
Yes, do it, please! That`s important for precise tournament testing

WHMoweryJr
27.07.2010, 05:58
http://www.mediafire.com/file/lz3801x9fk4aef1/IvanHoe-BetaWH_13.7z


Can somebody pls make some fast 64-bit builds
since my 32-bit machine makes
only very very very slow 64-bit builds...

In Arena loading bases is done once,
very easily,
and remembered forever for that engine,
by Arena.

halestorm
27.07.2010, 07:50
Someone said that they have never found someone to beat DR in long time controls. I asked what is the beginning of long. The answer was 10 minutes. So...

10+0, 50 games

1: IvanHoe 63Mod5-turbo-x64-i7 26.5/50 18-15-17 53%
2: Deep Rybka 4 SSE42 x64 23.5/50 15-18-17 47%


I7 920 at 3.5 ghz, 6 gb RAM, Window 7 64-bit

Bases ON for both (of course)
Book ON
Hash 1024

Games attached.

Rendruk
27.07.2010, 09:26
I'd call long 15+, probably 20. cegt goes 40/4 40/20

Peterpan
27.07.2010, 11:59
Someone sent me a private message here so i though while i replied to it,might as well post my latest compile here as well.

http://www.speedyshare.com/files/23540696/Ivanhoe-BPP0.1Intel64.rar :D:sm36:
http://www.speedyshare.com/files/23540913/Ivanhoe-BPP0.1AMD64.rar :D:sm36:
http://www.speedyshare.com/files/23540976/Ivanhoe-BPP0.132bits.rar :D:sm36:
It includes Intel sse3 and sse4.2 compile.
It also now includes AMD and AMD popcnt compile which i quickly made now.
It also now includes AMD and Intel 32 bits compiles
have fun :)

Noobster
27.07.2010, 12:04
Someone sent me a private message here so i though while i replied to it,might as well post my latest compile here as well.

http://www.speedyshare.com/files/23540696/Ivanhoe-BPP0.1Intel64.rar :D:sm36:

It includes a sse3 and sse4.2 compile.

have fun :)

What is it?

Rendruk
27.07.2010, 12:17
WTF happened to the moonrider chess page? Why does no one care about these important matters but me:P!

BTO7
27.07.2010, 14:24
http://www.mediafire.com/file/lz3801x9fk4aef1/IvanHoe-BetaWH_13.7z


Can somebody pls make some fast 64-bit builds
since my 32-bit machine makes
only very very very slow 64-bit builds...

In Arena loading bases is done once,
very easily,
and remembered forever for that engine,
by Arena.

Ok something is wrong. I'm canceling this tournament as Ivan is hanging up for some reason. I walked away from computer came back couple hours later and this is what i find...

»»»»Game 1 of 20: IvanHoe-BetaWH_13 - Deep Rybka 4 w32««««
Game end: 0-1 {0-1 White forfeits on time}
Last game won by Deep Rybka 4 w32 with Black in 2 moves

»»»»Game 1 of 20: Stockfish-18-32-ja - IvanHoe-BetaWH_13««««
Game end: 1-0 {1-0 Black forfeits on time}
Last game won by Stockfish-18-32-ja with White in 3 moves

»»»»Game 2 of 20: Deep Rybka 4 w32 - IvanHoe-BetaWH_13««««
Game end: 1-0 {1-0 Black forfeits on time}
Last game won by Deep Rybka 4 w32 with White in 2 moves

»»»»Game 2 of 20: IvanHoe-BetaWH_13 - Stockfish-18-32-ja««««
Game end: 0-1 {0-1 White forfeits on time}
Last game won by Stockfish-18-32-ja with Black in 5 moves

and so on. This was 5 min tourney and i have a core2duo 32 bit ...any ideas? I made no adjustments to the engine other then pawn hash 64.

Regards
BT

sirabc
27.07.2010, 15:27
From http://www.open-chess.org/viewtopic.php?f=7&t=470&start=10#p4474
kingliveson » Mon Jul 26, 2010 10:53 am
999953 was re-uploaded a few minute ago and will not be touched again. Everything is now back to default source after playing around. Normally I try not to mess with parts of the code that affect engine play.

Interesting statement posted on Ippolit website that there are secret modification in my builds. This is not quite the case. It is true that the builds do have slight modification, but in most cases, these are code already in the source -- disabled/enabled by default.

Search, Eval, Material Weight... are not tampered with -- many others already do that. The goal has always been to have something as close as possible to the original source, and as intended by the developers. Here are things I do and have done with the builds:

- In the earlier versions, Smooth Scaling (introduced to Stockfish by Dann Corbit) was added as UCI option up to version 999963. Although UCI option for 999963 is there, I actually forgot to enable the code -- later discovered when playing around with the source. It was a copy/paste job from RobboLito -- I believe Sentinel and Kranium added it.

- Auto CPU cores detection is in the source code but disabled by default. So it is enabled and called in main. TitanicCPUsSizings renamed to Threads.

- Pawn Hash is set to 2 MB by default in the source code. Left it at default in some builds while others use 4 MB or 8 MB.

- Send currmove is off by default in the source code, but on by compile.

- Added POPCNT code for 32-bit Windows builds. But of course you should be running a 64-bit OS if your CPU supports POPCNT.

- AllowInstantMoveFromHash default is off by compile as of 999954 if am not mistaken.

- After running 3000 games, 2 additional time management adjustments disabled by default in the source code are enabled in 999953 (this is the file that was just re-uploaded).

- 999953 should also now display cpuload correctly -- copy/pasted code from RobboLito.

- DoHashFull default is off by compile as of 999953. It has no use in game play.

All in all, I think more credit is being given me than is deserved.

Thanks,

Franklin

Get it here. http://chess.cygnitec.com/engine/ivanhoe/

BTW, I crc the exe files and they are different from the ones posted on Sunday.

Ahmed
27.07.2010, 15:28
Ahmed, add autoloading of Robbobases to your new compilation, please!
why you need this and you can now make { LoadRobboTripleBases , UnloadRobboTripleBases and RegisterRobboTotalBases , DeregisterRobboTotalBases } as you want in any time ,i think this is better
Ahmed what is this Ivanhoe 9.53b x64 and where can i find it?
http://chess.cygnitec.com/engine/ivanhoe/ :sm105:

Sandi
27.07.2010, 15:28
Skulltrail QX 9775 Yorkfield (8X4090 MHz)
Fritz Benchmark: Relative speed = 38,58. kN/s = 18.387.
Time control 10' (equivalent to a standard quad : 25' )
Hash 512.GUI Fritz 12. Ponder off. Nalimov tablebases 3-4-5 + Robbo Triple Bases. book:IPPOLIT-7moves.ctg.

Deep Rybka 4 - IvanHoe 9.53b x64 :
28.5 - 21.5 . 57% - 43% ( +12 -5 =33)
In long time control to 8 threads with full speed the KLO compile has tactical
problems.

The 50 games:

jc.m
27.07.2010, 17:08
This was 5 min tourney and i have a core2duo 32 bit ...any ideas? I made no adjustments to the engine other then pawn hash 64.


Yes, there is a bug where Ivanhoe 99953 enters dead mode, and it's not able to quit it.
Vlad0 also encountered the problem with 64 bits, but at this moment, we have no idea where the bug is.
I've activated the Source Analysis when compiling the sources with VS2010, and it discovered some small glitches, but none that could explain this behavior.

BTW, I just reinstalled my computer from Vista 32 bits (and VS2008) to Windows 7 64 bits (and VS2010), so I'll probably be able to test 64 bits soon :D
Still reinstalling software right now...

Ipman
27.07.2010, 17:21
Hi jc.m ,you gonne enjoy win 7 x64 ..put it on Best Performance!!

So,does this mean you gonne later also make 64bit compiles for us!?

Ipman.

TCH
27.07.2010, 17:27
why you need this and you can now make { LoadRobboTripleBases , UnloadRobboTripleBases and RegisterRobboTotalBases , DeregisterRobboTotalBases } as you want in any time ,i think this is better
For tournaments with robbobase on

Ahmed
27.07.2010, 17:30
IvanHoe T53A
IvanHoe T53A w32
IvanHoe T53A x64
IvanHoe T53A POPCNT
IvanHoe T53A SSE4.2
http://www.speedyshare.com/files/23543951/IvanHoe_T53A.rar
Good luck

jc.m
27.07.2010, 19:18
So,does this mean you gonne later also make 64bit compiles for us!?


Yes, and I'll commit the project on the SVN, so everybody could be able to do their own builds.
However, the 9999953 trunk is not very stable, or more exactly in my case, it loses a lot on time when playing 2 minutes games.

Yesterday, I wrote several new fast PopCount routines in assembler (I have 7 routines for that !), since I don't have a SSSE4 computer. I'll probably commit the code in the trunk too.

Sandi
27.07.2010, 19:31
Intel Xeon X5355 (8X2660 MHz).
Time: 10'+0".
Hash 512, GUI Fritz 12,Ponder off,Tablebases 5 pieces + Robbo Triple Bases.
Book: IPPOLIT-7moves.ctg.

IvanHoe T55A (SSE3) x64 - Deep Rybka 4 : 16 -14 (+8 -6 =16)

WHMoweryJr
27.07.2010, 19:34
Engine Score St

1: Stockfish-18-32-ja 97.0/204 ···················· ··············

2: IvanHoe-BetaWH_13KDC120 21.5/34 0=01=110=11=====1001==01=11111=1=1
3: IvanHoe-BetaWH_13KDC105 19.5/34 ===1===01=0101====11=10=00==11==11
4: IvanHoe-BetaWH_13KDC75 18.5/34 ===111=01=010===01===1000100111=1=
5: IvanHoe-BetaWH_13KDC60 17.0/34 011001=11==0=00===11=10=1001=10=0=
6: IvanHoe-BetaWH_13KDC90 16.0/34 ==00010=1000100=11010==1=101001=11
7: IvanHoe-BetaWH_13KDC45 14.5/34 ==10010======01001=0=101==000==1=0

Tourney: IvanHoe-BetaWH_13 KingDangerCentis
Level: Blitz 1min/0.5sec
2-cores
64M hash
Titus.abk
E-Mail Address: WHMoweryJr@yahoo.com

KDC == KingDangerCentis == a new UCI to test.

Monotonically increasing to 120 {+noise},
so higher values need to be tested...

This is the Stockfish tuning
which is not optimal for Rybka nor Houdini...

immortal223
27.07.2010, 19:39
IvanHoe T55A
The best Ivan!

BTO7
27.07.2010, 19:46
Yes, there is a bug where Ivanhoe 99953 enters dead mode, and it's not able to quit it.
Vlad0 also encountered the problem with 64 bits, but at this moment, we have no idea where the bug is.
I've activated the Source Analysis when compiling the sources with VS2010, and it discovered some small glitches, but none that could explain this behavior.

BTW, I just reinstalled my computer from Vista 32 bits (and VS2008) to Windows 7 64 bits (and VS2010), so I'll probably be able to test 64 bits soon :D
Still reinstalling software right now...

Thanks Jc.....I was gonna add i had ponder on if that mattered any, but as long as you guys know whats up thats the main thing :).

Regards
BT

Ahmed
27.07.2010, 19:47
The best Ivan!
Thanks, anyway i will make test between :
IvanHoe T55A Vs IvanHoe T53A
and the result coming soon

никел
27.07.2010, 19:49
999952 UCI_OPTIONS: STATIC_WEIGHTING MOBILITY_WEIGHTING
Time options
New 100 move fix
stop in MultiPV fixed
Fix Robbo_mossa with DTC metric, effects "mainline" and direct
BLOCK.INDEX for RobboBases
More UCI bug fixes with complianing
EvalHash to UCI
Fact: Too large is bad (cache)! 2mb with LARGE 3% above 16mb
Upon no large: set SmallEvalHash (kb)


http://ippolit.wikispaces.com/file/view/IvanHoe-Beta-999952a-Beta.tar/154194553/IvanHoe-Beta-999952a-Beta.tar

BTO7
27.07.2010, 19:50
Also this might help ....Beta 52 !!
http://ippolit.wikispaces.com/file/detail/IvanHoe-Beta-999952a-Beta.tar

Regards
BT

immortal223
27.07.2010, 19:53
Thanks, anyway i will make test between :
IvanHoe T55A Vs IvanHoe T53A
and the result coming soon

http://lh3.ggpht.com/_UkRyFNqhRJc/TE7ysgpmUkI/AAAAAAAAAI0/a2mvsJBsZbM/screenshot5422828.jpg

Noobster
27.07.2010, 19:55
http://lh3.ggpht.com/_UkRyFNqhRJc/TE7ysgpmUkI/AAAAAAAAAI0/a2mvsJBsZbM/screenshot5422828.jpg

Cant be true :sm10:

immortal223
27.07.2010, 19:57
Cant be true

Just finished testing :to_become_senile:
55 by KLO also won my huge 87-55 tournament

Ipman
27.07.2010, 20:16
Yes, and I'll commit the project on the SVN, so everybody could be able to do their own builds.

So,if we have a source for a new compile..i have a core i7 system..then it will be possible to make my own popcnt or SSE4.2 version?!!
Are al the settings needed to make this compile for my system then automatic,because i think most people here will not know how to use these settings!

Ipman.

Sandi
27.07.2010, 20:17
Skulltrail Intel QX 9775 Yorkfield (8X4090 MHz)
Time 3'+0".Hash 128. GUI Fritz 12. Ponder off.Tablebases 5 pieces + Robbo Triple Bases. Book: HS-7moves.ctg.

IvanHoe-B PP-0.1 - Deep Rybka 4 :
26-24. 52%-48%. (+11 -9 =30)
Very strong compile to 8 threads.

The 50 games:

jc.m
27.07.2010, 20:29
So,if we have a source for a new compile..i have a core i7 system..then it will be possible to make my own popcnt or SSE4.2 version?!!
Are al the settings needed to make this compile for my system then automatic,because i think most people here will not know how to use these settings!

Ipman.

Yes, all the settings are set exactly as the original build by the Ippoliters.
In fact, I just read their makefile and convert it into a Windows makefile.

I'll commit the new 52 version on the trunk and test if the compile works.

However, to create a build yourself, you need:
1) Visual Studio 2010 or at least VS2008
2) Intel C++ compiler, and this one is not currently available for VS2010

I have not yet finished the IC++ automated build, but I could do it with the previous version. However, since Intel compilers prefer Intel processors, I doubt that a 64 bits build will be good for the AMD64.

Ahmed
27.07.2010, 20:31
http://lh3.ggpht.com/_UkRyFNqhRJc/TE7ysgpmUkI/AAAAAAAAAI0/a2mvsJBsZbM/screenshot5422828.jpg
okay, i think this test is Enough
Really, IvanHoe T55A is very strong :sm36:
Let's see what will happen in IvanHoe 52a

jc.m
27.07.2010, 20:33
okay, i think this test is Enough
Really, IvanHoe T55A is very strong :sm36:
Let's see what will happen in IvanHoe 52a

Not really.

In fact, it's version 53 that is very slow.
My own tests show that it lost in time 95% of the games.
And I had worst results, for example 10 games at 2 minutes all lost with 53 !

It seems that they fixed several bugs in 52.

Ahmed
27.07.2010, 20:40
In fact, it's version 53 that is very slow.

I noticed this
T55A and T53A is same Settings but T55A is very faster than T53A

jc.m
27.07.2010, 20:46
I noticed this
T55A and T53A is same Settings but T55A is very faster than T53A

Not necessarily but v53 tends to take its time in the last moves when there is no more time, instead of playing instantly.
Several times, it took 2 to 3 seconds to play when the time was 0, so it was losing with -1 or -2 remaining seconds.
Edit: You cannot imagine how this behavior is irritating !

Since I reinstalled my computer, I have to reinstall TortoiseSvn to commit the new sources in the trunk. I'll then check what has changed.

Ahmed
27.07.2010, 20:53
Not necessarily but v53 tends to take its time in the last moves when there is no more time, instead of playing instantly.
Several times, it took 2 to 3 seconds to play when the time was 0, so it was losing with -1 or -2 remaining seconds.
Edit: You cannot imagine how this behavior is irritating !

Since I reinstalled my computer, I have to reinstall TortoiseSvn to commit the new sources in the trunk. I'll then check what has changed.
i think IvanHoe 55 is better than IvanHoe 53 in the time used and the speed
anyway, Let's see what will happen in IvanHoe 52a

jc.m
27.07.2010, 21:29
i think IvanHoe 55 is better than IvanHoe 53 in the time used and the speed
anyway, Let's see what will happen in IvanHoe 52a

I already committed the version, and it seems to compile fine with my Nant script, except one problem:

[exec] SLAB_MEMORY.c(13) : error C2036: 'void *' : unknown size
[exec] SLAB_MEMORY.c(17) : error C2036: 'void *' : unknown size
[exec] SLAB_MEMORY.c(18) : error C2036: 'void *' : unknown size
[exec] SLAB_MEMORY.c(20) : error C2036: 'void *' : unknown
[exec] SLAB_MEMORY.c(43) : error C2036: 'void *' : unknown size

to fix that, just change the two lines:
static unsigned char* SLAB_ROOT_LOC = NULL;
static unsigned char* CURRENT_SLAB_LOC;
(you cannot do pointers' arithmetic on void* in C)

and it compiles perfectly :D
I'll post the 32 bits builds in a few minutes, and will finish the 64 builds in a few hours...

jc.m
27.07.2010, 21:42
I just uploaded the builds:

http://code.google.com/p/ivanhoebuilds/downloads/list

Grab the 999952a version.
Since I reinstalled my computer, I don't have IC++ compiler anymore, so only VS2010 builds now.
I'll install IC++ then try to do my first 64 bits build (but Vlad0 already told me what to do, it will be easy).

Rendruk
27.07.2010, 22:28
"Fact: Too large is bad (cache)! 2mb with LARGE 3% above 16mb"

I used to think that the reason they used a translator was just to be funny. . . but there are times to be funny and there are times to make enough sense to be understood. . .

I ask again, *why* did the moonrider homepage go down? Also, where can we get books now that it is down?

Tomcass
27.07.2010, 22:36
Yes, and I'll commit the project on the SVN, so everybody could be able to do their own builds.
However, the 9999953 trunk is not very stable, or more exactly in my case, it loses a lot on time when playing 2 minutes games.

Yesterday, I wrote several new fast PopCount routines in assembler (I have 7 routines for that !), since I don't have a SSSE4 computer. I'll probably commit the code in the trunk too.

Jc.m, this is very good news for our community!. Thank you very much for your effort!. :sm36:

Tom.

slankamen
27.07.2010, 23:11
http://rebornse.niceboard.net/forum.htm

Hi Rendruk,
this is a site where you can find a lot of opening books.Just register and you should be able to download a lot of opening books.
Looks like Moonrider was either tired of hustle with the book authors and the fights or simply lost the site.
Also on the new Openchess forum there is book section so this is another site for book download.

Vlad0
27.07.2010, 23:51
After the "sleeping bug" was localized (thanks to LeonardoVinci on
http://ippolit.wikispaces.com/message/view/Bugs/25846893 ) today abruptly arose
a big number of relatively stable 53 engines, alternative to KLO's. I also made one
(see attachment to my last blog on http://immortalchess.net/forum/blog.php?u=971 ).
I tested it this night in 200 blitz 1+0 games with KLO's 9.53b. The result was:

1 IvanHoe 53bU-x64 +41 +51/=119/-28 55.81% 110.5/198
2 IvanHoe 9.53b x64 -41 +28/=119/-51 44.19% 87.5/198

Both engines were with the same UCI options. In particular, Large pages were on,
but Total bases were off. Because the purpose of the test was just to confirm
my suspicion, that "sleeping bug" was appearing in the presence of Total bases only).
This was confirmed by the test (not a single "sleep" occured). Instead there were
2 cases of "no move" bug in 53bU, but they were overcompensated by 21 points of
being ahead of 9.53b :) for some reason.

Ipman
27.07.2010, 23:58
Yes Vlad0..we see your great compiles..we can click on it for download..but we don't get permission to download them.

So,somebody who has ,can download Vlad0 compile..we would be glad that we can download it here and test it!

Ipman.

jc.m
28.07.2010, 00:35
Wow, the VS2010 C++ compiler crashes when compiling pawn_eval.c for 64 bits...

Am I the only one with this problem ?

immortal223
28.07.2010, 00:47
Vlad0, эта твоя компиляция на 10-11% медленнее чем 55-ая от Ахмеда (POPCNT)

Noobster
28.07.2010, 00:51
Vlad0, эта твоя компиляция на 10-11% медленнее чем 55-ая от Ахмеда (POPCNT)

in english section we should speak english or Azaad will ban you :acute:

Sandi
28.07.2010, 01:05
I noticed this
T55A and T53A is same Settings but T55A is very faster than T53A

Hi,
8 threads with 8 phisical cores is another world!
The T53A is here nearly on the same level with T55A(SSE3) !!
Only T55A(noSSE) is clear the best.
Skulltrail Intel QX 9775 Yorkfield (8X4090 MHz).
Time 3'+0".Hash 128.GUI Fritz 12.Ponder off.Tablebases 5 pieces + Robbo Triple Bases. Book: HS-7moves.ctg.:

Deep Rybka 4 - IvanHoe T53A x64 : 25.5 - 24.5 .51% -49%.(+9 -8 =33)
Very close match!

The 50 games:

jc.m
28.07.2010, 01:29
I have my first 64 bits compiles, and their are 2 times faster than the old 32 bits !!!

I'll publish my VC+Intel 64 bits compiles in a few hours, once the Intel are stable enough...

Rendruk
28.07.2010, 02:10
http://rebornse.niceboard.net/forum.htm

Hi Rendruk,
this is a site where you can find a lot of opening books.Just register and you should be able to download a lot of opening books.
Looks like Moonrider was either tired of hustle with the book authors and the fights or simply lost the site.
Also on the new Openchess forum there is book section so this is another site for book download.

Thanks very much Slankamen! Least someone's paying attention:P I think knowing where to get books is very important and very hard to learn too, because when you do a search for books you see rybka forum, they don't tell you where, you see pioner, they don't tell you where, you see sedat, he has one old book to get but doesn't tell you where to get more, you see gladiator, they only have outdated books like uptown 1.2 instead of 2.1. . .

So it's hard to figure out. Then this forum doesn't update books either in its book section, at least not compared to moonriders page. So again, thanks!

jc.m
28.07.2010, 02:42
All builds now available !

32 or 64 bits, VC++ or IC++, POPCNT or not.
Grab the files here:
http://code.google.com/p/ivanhoebuilds/downloads/list
(999952a.zip)

I also committed the build tool and the vcprojx. It takes 10 minutes to create all these builds, and it's completely automated now.
Warnings:
- Intel C++ 11 builds crash when you force the full optimization (/Qipo or /GL). So I had to disable these.
- Visual C++ 2010 crashes when compiling from the vcprojx, but not from the command line.

Also, I don't have a SSE4 processor, so I cannot test the POPCNT versions. I hope they are stable.

Compilers: don't hesitate to use these sources and propose better compiler options (I put all the options I know were safe).

Please report problems here.

slankamen
28.07.2010, 02:44
Q9550-XP64 3.5Ghz, Blitz:10' 0


1 IvanHoe T53A.x64 3185 +6/-3/=21 55.00% 16.5/30
2 Deep Rybka 4 x64 3190 +3/-6/=21 45.00% 13.5/30

jc.m
28.07.2010, 03:02
Please report problems here.

When running the engine with "benchmark", they seem correct, but when running with Arena, all the 64 bits engine crash badly.
I'll investigate the problem.

Edit after some tests: the SSSE3 VC++ 64 bits works properly.
The code seems to crash because of the optimizations I set for IC++.
I'll try to find which optimization crash the code.

Second edit: Ok, found it !
IC++ builds crash when I keep the aliasing. Adding /Oa makes the builds work :D

yayo
28.07.2010, 05:28
Vlado can you give a link here to Ivanhoe53bu please?

jc.m
28.07.2010, 05:36
64 bits builds seem stable now:
http://code.google.com/p/ivanhoebuilds/downloads/list
Grab the file 999952a-64bits.zip.

CS29797
28.07.2010, 05:39
All builds now available !

32 or 64 bits, VC++ or IC++, POPCNT or not.
Grab the files here:
http://code.google.com/p/ivanhoebuilds/downloads/list
(999952a.zip)

Mediafire Mirror (http://www.mediafire.com/download.php?aq7744q2a5ofsh1)

slankamen
28.07.2010, 06:52
Still looking for the magic version that is constantly going to win against Rybka.

Hardware: Intel(R) Core(TM) i7 960 3.20GHz8x @ 4.005 GHz with 4,096 MB Memory
Operating system: Microsoft Windows XP 64 Bit Professional Service Pack 2 (Build 3790)
Fritz Benchmark:
Speed: 24.98
KNS: 11988
GUI: CB Rybka 3
Book: Perfect 2009 10 moves
Hash: 256
RB and TB: ON
Ponder: OFF

Blitz:10' 0

1 Deep Rybka 4 SSE42 x64 3190 +4/-2/=14 55.00% 11.0/20
2 IvanHoe 53bU-x64 3195 +2/-4/=14 45.00% 9.0/20

jeandis
28.07.2010, 07:22
64 bits builds seem stable now:
http://code.google.com/p/ivanhoebuilds/downloads/list
Grab the file 999952a-64bits.zip.

this engine crashed

halestorm
28.07.2010, 08:01
Someone said no one beats DR4 in long time controls. I asked what the shortest time control was to "long." They said 10 minutes. I posted 10 minutes yesterday. Someone then said 15+0.

Here is 20+0, 40 games.


Deep Rybka 4 SSE42 x64 19.5/40 9-10-21 49%
IvanHoe 63Mod5-turbo-x64-i7 20.5/40 10-9-21 51%


bases ON (both, of course)
book ON
I7 920 oc to 3.5 ghz, 6 gig ram


Games attached.

slankamen
28.07.2010, 08:26
I would like to make a comment as a person who likes to do the engine testing.
Recently we have been bombarded by different versions and compiles that is very difficult to follow and even more difficult to test.
Just the latest version T52 in the folder has 19 different compiles.10 of them are 64 bit and 9 32 bit versions.I have tried one version on i7 that crashed so I had to remove it from the engine list and was lucky with the second version that was working OK.
I am very much grateful to all the compilers who are participating in this forum and releasing their work for us to test but with so many versions available how is one going to know which version to test.They are being released to be tested but I wander if some preliminary testing was done to be sure that all the versions are working.
I have now 4 computers at my disposition for engine testing but even that would be totally inadequate considering the number of engines available .
Instead of working together different compilers release their engines sometimes in the same day so one finds itself in dilemma which engine to test first.Than corrections are being made and new compiles released complication already bad situation.
We don't have time to determine the strength of one version and new one is released.
On top of that there are other programs coming from different programmers not related to our site that are also worth doing the testing.
So, the question is how to solve this problem.
Could it be possible to release as few engines as possible that first work OK and to be simple as possible to.Not the list of so many but version for 32 bit as well as basic 64 bit versions including those for i7 computers. Three to to four engines in one folder.That was the way engines were released before but than their number started to grow with different bells and whistles.
Testing of the engines should be fun but what was going on recently has made the situation complicated.
I would like to hear what other people who do the testing are going to say about the present situation and to express their suggestions.

Rendruk
28.07.2010, 08:31
I would like to make a comment as a person who likes to do the engine testing.
Recently we have been bombarded by different versions and compiles that is very difficult to follow and even more difficult to test.
Just the latest version T52 in the folder has 19 different compiles.10 of them are 64 bit and 9 32 bit versions.I have tried one version on i7 that crashed so I had to remove it from the engine list and was lucky with the second version that was working OK.
I am very much grateful to all the compilers who are participating in this forum and releasing their work for us to test but with so many versions available how is one going to know which version to test.They are being released to be tested but I wander if some preliminary testing was done to be sure that all the versions are working.
I have now 4 computers at my disposition for engine testing but even that would be totally inadequate considering the number of engines available .
Instead of working together different compilers release their engines sometimes in the same day so one finds itself in dilemma which engine to test first.Than corrections are being made and new compiles released complication already bad situation.
We don't have time to determine the strength of one version and new one is released.
On top of that there are other programs coming from different programmers not related to our site that are also worth doing the testing.
So, the question is how to solve this problem.
Could it be possible to release as few engines as possible that first work OK and to be simple as possible to.Not the list of so many but version for 32 bit as well as basic 64 bit versions including those for i7 computers. Three to to four engines in one folder.That was the way engines were released before but than their number started to grow with different bells and whistles.
Testing of the engines should be fun but what was going on recently has made the situation complicated.
I would like to hear what other people who do the testing are going to say about the present situation and to express their suggestions.

Solution: Have a main compiler who takes suggestions from the other compilers.

In other words; have the complier who makes the fastest complies because of his computer hardware/software (I.e. 64 bit, intel, VS2010, etc. . . )

Make a compile, but have some actual communication between the usual people, that way the good ideas can all be COMBINED into one set!

halestorm
28.07.2010, 08:42
We don't have time to determine the strength of one version and new one is released.

This mirrors my words from a few days back. I have many computers at my disposal, including 2 I7s and many 32-bit. But to test takes time. We want to help, but don't know how to keep up with the builds. Perfect, Slank!

ps not for slankamen: questions about opening books that have nothing to do with ivan probably belong in a forum for opening books. we pay attention.

BTO7
28.07.2010, 09:42
I would like to make a comment as a person who likes to do the engine testing.
Recently we have been bombarded by different versions and compiles that is very difficult to follow and even more difficult to test.
Just the latest version T52 in the folder has 19 different compiles.10 of them are 64 bit and 9 32 bit versions.I have tried one version on i7 that crashed so I had to remove it from the engine list and was lucky with the second version that was working OK.
I am very much grateful to all the compilers who are participating in this forum and releasing their work for us to test but with so many versions available how is one going to know which version to test.They are being released to be tested but I wander if some preliminary testing was done to be sure that all the versions are working.
I have now 4 computers at my disposition for engine testing but even that would be totally inadequate considering the number of engines available .
Instead of working together different compilers release their engines sometimes in the same day so one finds itself in dilemma which engine to test first.Than corrections are being made and new compiles released complication already bad situation.
We don't have time to determine the strength of one version and new one is released.
On top of that there are other programs coming from different programmers not related to our site that are also worth doing the testing.
So, the question is how to solve this problem.
Could it be possible to release as few engines as possible that first work OK and to be simple as possible to.Not the list of so many but version for 32 bit as well as basic 64 bit versions including those for i7 computers. Three to to four engines in one folder.That was the way engines were released before but than their number started to grow with different bells and whistles.
Testing of the engines should be fun but what was going on recently has made the situation complicated.
I would like to hear what other people who do the testing are going to say about the present situation and to express their suggestions.

I agree and thats why i have been holding back of late. Since 63 things have got hectic. I believe WHM is working in this direction with a few of the guys. So I'm kinda waiting to see what that team comes up with. The new sources are not claiming elo but only bug fixes and with that in mind a few programmers have said Ivan is close to optimal. So a team compiling the code with all tricks from each avail is gonna give us all the best ...most stable engine. I see WHM starting other treads and trying to build a version with all the best tricks into one ...so that eventually when a new release comes out the new additions to the code can be just added to a already stable compile thats fast with out starting from scratch every compile. At least this is what I think hes trying to do. To me testing is looking for more elo and better scores. If there is no elo added to compile then we already know how strong it is and not too much reason to test. I wanna test the best of all ideas version by a team and so far it only seems WHM is willing to put enough ego to the side to share. I would rather a team effort compile to test myself.

Regards
BT

TCH
28.07.2010, 10:05
Recently we have been bombarded by different versions and compiles that is very difficult to follow and even more difficult to test. Just the latest version T52 in the folder has 19 different compiles.10 of them are 64 bit and 9 32 bit versions.
I don't think that it's a big problem, on the contrary, i prefere to have a choice. If I know that on my computer 53 SSE4.1 was stronger than any other 32-bit compilation, I can assume that 52 SSE4.1 will be also stronger than others. In order to verify that it's possible to run the tournament with an ultrashort control, which does not take much time. Long control here is not necessary, because we have to compared only the speed of engines.

dollyman
28.07.2010, 10:52
ok , waiting for new compil

good luck

Captain
28.07.2010, 15:26
64 bits builds seem stable now:
http://code.google.com/p/ivanhoebuilds/downloads/list
Grab the file 999952a-64bits.zip.

I've tried several of these compiles. But i don't get as high nodes per second as previous versions of ivanhoe. With these compiles i get only ~3000 nodes/sec with versions ivanhoe 63 i get ~10000 nodes/sec.

What's the cause? :sm33:

Hardware: Intel(R) Core(TM) i7 920 8x2,7GHz @ 4.0 GHz.
Memory for Engine: 2 MB.
Operating system: Microsoft Windows Vista 64 Bit

funchess
28.07.2010, 15:33
after long time ego, what ivanhoe version is now the strongest ?

TCH
28.07.2010, 15:36
I'm not sure than there is any progress in last builds. For example, in a small match with ultrashort TC 15 sec/game without robbobases:
IvanHoe T63D w32 : Ivanhoe999952_jc.m_32_SSE4.1_IC 107.5:92.5 +63=89-48
IvanHoe T63D is stronger with confidence 92,2%

Rendruk
28.07.2010, 15:59
Halestorm: I think it's cool you tested the games at 20 minutes like I mentioned. That takes a lot of time and we appreciate it.

This makes me think of Slankaman who apparently has 4 computers devoted to constantly testing 10 minute games.

Which is Fu*king insane my man!

How about you give me some money, I'll do some tests too:P

Ipman
28.07.2010, 16:02
jc.m ,thanks for all the compiles!

I always run first a benchmark ,using "go depth 22" to see which one is the fastest one and use that one in my tests.
Now these x64 compiles are setted to 1 core..how i set it to 4 cores?.."mt=4" don't works ,is there a other way or it has to make how you have build these compiles!

Ipman.

slankamen
28.07.2010, 16:32
Intel(R) Core(TM)2 Quad Q9550 2.83GHz4x @3500 MHz 4,096 MB Memory
Microsoft Windows XP 64 Bit Professional Service Pack 2 (Build 3790)
Fritz Benchmark:
Speed: 20.90
KNS: 10032
GUI: CB Rybka 3
Book: Perfect 2009 10 moves
Hash: 256
RB and TB: ON
Ponder: OFF

Blitz:10' 0


1 IvanHoe-Beta version 999952a- 3185 +10/-4/=26 57.50% 23.0/40
2 Deep Rybka 4 x64 3190 +4/-10/=26 42.50% 17.0/40

:sm36:

Ipman
28.07.2010, 16:35
Blitz 5m Core i7 920 @3.87Ghz , Blitz 5 0

Ivanhoe-B PP-0.1 2800 - Houdini 1.02 x64 POPCNT 4_CPU 2800 5.0 - 5.0 +1/-1/=8 50.00%
Ivanhoe-B PP-0.1 2800 - IvanHoe-Beta version 999952a- 2800 6.0 - 4.0 +2/-0/=8 60.00% -> was the SSE4.1 x64
Ivanhoe-B PP-0.1 2800 - Ivanhoe-B57d_whm01_w64 2800 5.0 - 5.0 +0/-0/=10 50.00%

Ipman.

halestorm
28.07.2010, 16:41
4+2

32-bit Win7, 2gb RAM
Aquarium interface, Narrow book, bases OFF


Deep Rybka 4 52/100 +27/=50/-23
Ivan T53A w32 48/100 +23/=50/-27

slankamen
28.07.2010, 16:43
AMD Phenom II 1090T.x64 , Blitz:10' 0

12345678901234
1 Deep Rybka 4 x64 3185 ½1½½½½½1111101 10.0/14
2 IvanHoe-Beta version 999952a- 3195 ½0½½½½½0000010 4.0/14

This version had multiple loses on time on AMD unlike the version who did so well on Quad 9550.The plan was 40 games but the engine stopped the match at the 15th game so I stopped the agony.

halestorm
28.07.2010, 17:07
this engine crashed

Ivanhoe64_SSE4.1_POPCNT_IC is flawed on my I7 system.
Consistently 30-40 out of 100 in STS testing.
Shame, since it has high nps.

calu
28.07.2010, 17:37
So far it seems like the king is back!!!
IvanhoeB PP0.1 is the most INCREDIBLE engine I have ever seen man! And I have seen and studied many of them as everybody in this forum did.
IvanhoeB PP0.1 It just dominates on the board with those magnificent plans, both strategies and tactics are coming as natural as if an adult tried to outsmart a kid during a chess game… I don’t know man… I am astonished with its performance.
Right now I run a blitz tournament 15’+10” between Ivanhoe B PP0.1 and rybka4 and so far the score is unbelievably high for Ivanhoe, +11 =18 -1. Both engines are using the same book-storm- until the 8th move.
I just have to give credits to everybody who managed to achieve such a good combination of strategic and tactical engine like Ivanoe B PP0.1
BRAVO!!!

seva_shilon
28.07.2010, 18:00
Сompilation IvanHoe 52
http://chess.cygnitec.com/engine/ivanhoe/IvanHoe999952a.7z

Ahmed
28.07.2010, 18:02
IvanHoe T52A
IvanHoe T52A w32
IvanHoe T52A x64
IvanHoe T52A SSE4.2
IvanHoe T52A POPCNT
http://www.speedyshare.com/files/23559699/IvanHoe_T52A.rar
New 100 move fix
stop in MultiPV fixed
Fix Robbo_mossa with DTC metric, effects "mainline" and direct
BLOCK.INDEX for RobboBases
More UCI bug fixes with complianing
EvalHash to UCI
Fact: Too large is bad (cache)! 2mb with LARGE 3% above 16mb
Upon no large: set SmallEvalHash (kb)
Good luck

slankamen
28.07.2010, 18:27
Can somebody be so kind to post a download site for the version IvanhoeB PP0.1

Ipman
28.07.2010, 18:31
Slankamen..see post 3118 on page 156!!

You can choose 3 versions!

Ipman.

jc.m
28.07.2010, 18:50
Hi and thanks for the feedback !

I'll try to answer to everybody in a single answer, please take time to read the answers, they might be interesting for the future...

this engine crashed

Sorry, but since I posted a lot of versions, this message is absolutely worthless.
Which version crashed ? Just tell me the executable's name.
When I do my tests, I take always the most interesting compile, in the following order:
64 bits > 32 bits
and
SSE4 > SSSE3 > SSE2 > SSE > plain (no SSE)

When the top version doesn't work, take the next one.


Just the latest version T52 in the folder has 19 different compiles.10 of them are 64 bit and 9 32 bit versions.I have tried one version on i7 that crashed so I had to remove it from the engine list and was lucky with the second version that was working OK.
I have now 4 computers at my disposition for engine testing but even that would be totally inadequate considering the number of engines available .


All the engines I propose are THE SAME, so there is no need to test all these versions as if they were different.
Just take the version that is the faster for your hardware.
If a version doesn't work, just take a version below, as I explained above.


In other words; have the complier who makes the fastest complies because of his computer hardware/software (I.e. 64 bit, intel, VS2010, etc. . . )


Since I opened the sources of my tools to compile the engines, I'll gladly give access to anyone interested in improving the builds.
I'd like to add something:
my goal is to create minimal builds from the original sources, and this is not the approach used by most compilers, who change the code to fix bugs, etc...
I don't intend to patch Ivanhoe, I just try to provide executables as near as possible as the original Linux versions.
Anyway, I'm not an expert in tuning compilers' options, so if you have better settings, I'll put them in the compile immediately.


I wanna test the best of all ideas version by a team and so far it only seems WHM is willing to put enough ego to the side to share. I would rather a team effort compile to test myself.


This is not a direction I'd like to take.
If compilers have tricks to improve the engine, it should be added directly in Ivanhoe's original version, not in the Windows' compiles.
First, having a Windows version stronger than the Linux one is not fair for the original version, which was created to share a strong open-source engine.
Second, it's a hell to maintain the upgrades from version to version.
If you have noticed, the Decembrists released a lot of versions these last months.
If we have to fork some functions, it will take a lot of time to upgrade to the latest version at every iteration.
So, when you have an improvement, please submit it to the Decembrists !
BTW, I'll post the new POPCNT functions soon here, I hope the Decembrists will use them for their builds.


I don't think that it's a big problem, on the contrary, i prefere to have a choice. If I know that on my computer 53 SSE4.1 was stronger than any other 32-bit compilation, I can assume that 52 SSE4.1 will be also stronger than others.

Exactly !
The difference between the versions are the nodes per second.
Since the C++ compilers have various bugs, I cannot predict which one will be stable.
In theory, a C++ compiler should always generate stable code, but it's not currently possible. In one day, I experienced several kinds of crashes, just by building the engines !


I've tried several of these compiles. But i don't get as high nodes per second as previous versions of ivanhoe. With these compiles i get only ~3000 nodes/sec with versions ivanhoe 63 i get ~10000 nodes/sec.
What's the cause? :sm33:


Frankly, I HAVE NO IDEA.
I've an Intel dual core, and get 1 million of nodes per second on a single core.
Also, as I mentioned above, some C++ compilers tend to crash because of the optimizations. I had to reduce the optimization to get stable Intel C++ compiles.

I'm not sure than there is any progress in last builds. For example, in a small match with ultrashort TC 15 sec/game without robbobases:
IvanHoe T63D w32 : Ivanhoe999952_jc.m_32_SSE4.1_IC 107.5:92.5 +63=89-48
IvanHoe T63D is stronger with confidence 92,2%

No, this doesn't prove anything.
It just proves that 999952 always manages badly its time.
The builds of various compilers here tried to fix the time-management problems, but I don't apply such trick: I just compile from the original sources.
Please report the problem to Decembrists.


I always run first a benchmark ,using "go depth 22" to see which one is the fastest one and use that one in my tests.
Now these x64 compiles are setted to 1 core..how i set it to 4 cores?.."mt=4" don't works ,is there a other way or it has to make how you have build these compiles!


"go depth 22" is as useless as "benchmark".
The only way to know which compile is the fastest is to test it in a real game.
In my case, I like to see Ivanhoe compete against Ruffian, since I use Arena.
I set 1 min + 1s, and the games are fun to watch.
To use multi-core, you have to type:
setoption name TitanicMode value true
setoption name TitanicCPUsSizings value 4


Intel(R) Core(TM)2 Quad Q9550 2.83GHz4x @3500 MHz 4,096 MB Memory

1 IvanHoe-Beta version 999952a- 3185 +10/-4/=26 57.50% 23.0/40
2 Deep Rybka 4 x64 3190 +4/-10/=26 42.50% 17.0/40

:sm36:

Nice !

AMD Phenom II 1090T.x64 , Blitz:10' 0

12345678901234
1 Deep Rybka 4 x64 3185 ½1½½½½½1111101 10.0/14
2 IvanHoe-Beta version 999952a- 3195 ½0½½½½½0000010 4.0/14

This version had multiple loses on time on AMD unlike the version who did so well on Quad 9550.The plan was 40 games but the engine stopped the match at the 15th game so I stopped the agony.

Yes, I encountered similar problems.
Sometimes, it loses badly on time, I have no idea why.

Ivanhoe64_SSE4.1_POPCNT_IC is flawed on my I7 system.
Consistently 30-40 out of 100 in STS testing.
Shame, since it has high nps.

Weird, can you confirm it with another version, like a 32 bits one ?
About the POPCNT version, I don't know if it works correctly, since I have only SSE3.

Lupus
28.07.2010, 19:21
Why Ivanhoe T52A can load no Bases???
:sm33:

halestorm
28.07.2010, 20:04
IvanhoeB PP0.1 is the most INCREDIBLE engine I have ever seen man!

Early testing on my end does not confirm:


T/C 5+0
Ivanhoe-BPP0.1sse3 -3/+10/=4 against DR4


Playing 250 games, so we'll see if things even out. But so far, I cannot say I am impressed. And I want to be! (where *is* the magic engine that beats DR4 consistently?)

WHMoweryJr
28.07.2010, 20:15
jc.m ,thanks for all the compiles!

I always run first a benchmark ,using "go depth 22" to see which one is the fastest one and use that one in my tests...

Ipman.


This is one of many observables - which are all needed,
not discounting any,
thanks for the post
and please post your
go depth 22
list of engines
which give us an idea of speed.

Then please correlate that
with measured playing ELO
in a table

Engine ELO Speed W-L-T


Thanks;
Bill

P.S. The correlation coefficient
between ELO and go depth 22 speed
will be much closer to 1 than 0.5
and way above 0.

P.P.S. Pls just ignore
the negative criticisms of the negative,
as they serve no purpose but to annoy,
thnx.

P.P.P.S. And yes there are builds that are fast early but slow late {compared to other builds}

P.P.P.P.S. And yes there are builds that are slow early but fast late {compared to other builds}

P.P.P.P.P.S. Thatsawhy i said "idea",
and i know of no builder
that would build a real fast startpos engine
at the expense of speed late in the game,
since all are trying to improve ELO.

checkmate4u
28.07.2010, 20:17
i too confirm that ivan b pp1 is not doing good infact it lost to ivan 5.3 b also not a single win for pp version of 32 bit

Ipman
28.07.2010, 20:39
Here still busy testing PP's compile IvanHoe-BPP0.1 and not loose a single match yet against 6 engines now..number 7 is running against IvanHoe 9.52b x64

Results will come later...

Ipman.

Lupus
28.07.2010, 20:50
Why Ivanhoe T52A can load no Bases???

Ipman
28.07.2010, 22:04
A update..still not loose yet!

Blitz 5m Core i7 920 @3.87Ghz , Blitz 5 0

Ivanhoe-B PP-0.1 2800 - Ivanhoe-B57d_whm01_w64 2800 5.0 - 5.0 +0/=10/-0 50.00%
Ivanhoe-B PP-0.1 2800 - Houdini 1.02 x64 POPCNT 4_CPU 2800 5.0 - 5.0 +1/=8/-1 50.00%
Ivanhoe-B PP-0.1 2800 - IvanHoe-Beta version 999952a- 2800 6.0 - 4.0 +2/=8/-0 60.00%
Ivanhoe-B PP-0.1 2800 - Houdini 1.03a x64 POPCNT 4_CP 2800 5.0 - 5.0 +1/=8/-1 50.00%
Ivanhoe-B PP-0.1 2800 - IvanHoe 9.53b x64 2800 5.0 - 5.0 +0/=10/-0 50.00%
Ivanhoe-B PP-0.1 2800 - IvanHoe-BetaWH_14_w64SSSE3 2800 5.5 - 4.5 +1/=9/-0 55.00%
Ivanhoe-B PP-0.1 2800 - IvanHoe 9.52b x64 2800 5.0 - 5.0 +1/=8/-1 50.00%

Ipman.

jc.m
28.07.2010, 22:18
Why Ivanhoe T52A can load no Bases???

Because you have to configure the engine to load the bases.
Sorry, I don't know how to do that with Arena.

But manually, you can test that this works as follows:

- double-click on an executable
- type:
setoption name RobboTripleBaseDirectory value C:\WHEREYOUSTOREYOURFILES

it should say info string Optional...
- then type:
setoption name LoadRobboTripleBases
it will display all the files that it loads.

jc.m
28.07.2010, 22:30
P.P.P.P.P.S. Thatsawhy i said "idea",
and i know of no builder
that would build a real fast startpos engine
at the expense of speed late in the game,
since all are trying to improve ELO.

Personally, I'm testing without using an opening book and without endgame tables, and Ivanhoe is tuned for them.
For example, when Ivanhoe is out of the opening book (aka after having played mecanically its first moves), it's starting to deep-compute, in order to get a grasp about the position.
As I test without opening book, Ivanhoe starts by taking 4 or 5 seconds for its first move, even when I run a one minute tournament.
In fact, it has no opening book knowledge at all, and this is pretty disturbing when you watch it play. For example, it tends to place its bishop in d3, but without moving its pawn from d2, which is a poor move.

About the speed of the engine, there is no solution.
You have to test the engines manually.
This is because we want to have the maximal speed, and this comes with multiple executables.

Finally, I'm still noticing some problems when running "TitanicMode" and "benchmark". Some times, it remains freezed during one or 2 seconds, without any reason (and its final nodes per second is 300000, instead of 1 million). I'm pretty sure that this bug is related to the problem that Vlad0 spent so much time fixing, and probably is the cause of the losses in time when blitzing.

WHMoweryJr
28.07.2010, 22:56
Here are some nps measurements at startpos.

======================================== ===============
go depth 22
=========

O1 Qx

1.IvanHoe-BetaWH_12_w64
info multipv 1 time 5788 nodes 44955765 nps 7767000 score cp 13 depth 20 pv e2e4
e7e5 b1c3 g8f6 g1f3 b8c6 f1c4 f8c5 d2d3 e8g8 e1g1 d7d6 c1e3 c5e3 f2e3 c8e6 c4e6
f7e6 d1d2 d8d7 c3b5 a7a6
info multipv 1 time 5788 nodes 44955765 nps 7767000 score cp 13 depth 20 pv e2e4
e7e5 b1c3 g8f6 g1f3 b8c6 f1c4 f8c5 d2d3 e8g8 e1g1 d7d6 c1e3 c5e3 f2e3 c8e6 c4e6
f7e6 d1d2 d8d7 c3b5 a7a6
bestmove e2e4 ponder e7e5

2.IvanHoe-BetaWH_12_w64SSE2
info multipv 1 time 4603 nodes 35092905 nps 7623000 score cp 13 depth 20 pv e2e4
e7e5 g1f3 b8c6 f1c4 f8c5 b1c3 g8f6 e1g1 d7d6 d2d3 e8g8 c1e3 c5e3 f2e3 c8e6 c4e6
f7e6 d1d2 d8d7 c3b5 a7a6
info multipv 1 time 4603 nodes 35092905 nps 7623000 score cp 13 depth 20 pv e2e4
e7e5 g1f3 b8c6 f1c4 f8c5 b1c3 g8f6 e1g1 d7d6 d2d3 e8g8 c1e3 c5e3 f2e3 c8e6 c4e6
f7e6 d1d2 d8d7 c3b5 a7a6
bestmove e2e4 ponder e7e5

3.IvanHoe-BetaWH_12_w64SSE3
info multipv 1 time 5835 nodes 44612285 nps 7645000 score cp 13 depth 20 pv e2e4
e7e5 g1f3 b8c6 b1c3 g8f6 f1c4 f8c5 e1g1 e8g8 d2d3 d7d6 c1e3 c5e3 f2e3 c8e6 c4e6
f7e6 d1d2 d8d7 c3b5 a7a6
info multipv 1 time 5835 nodes 44612285 nps 7645000 score cp 13 depth 20 pv e2e4
e7e5 g1f3 b8c6 b1c3 g8f6 f1c4 f8c5 e1g1 e8g8 d2d3 d7d6 c1e3 c5e3 f2e3 c8e6 c4e6
f7e6 d1d2 d8d7 c3b5 a7a6
bestmove e2e4 ponder e7e5

4.IvanHoe-BetaWH_12_w64SSE4.1
info multipv 1 time 5741 nodes 44534886 nps 7757000 score cp 14 depth 20 pv e2e4
e7e5 b1c3 g8f6 g1f3 b8c6 f1c4 f6e4 c3e4 d7d5 c4d3 d5e4 d3e4 c6e7 e4d3 f7f6 e1g1
d8d5 d1e2 c8f5 d3c4 d5e4 e2e4 f5e4 d2d3 e4f3 g2f3
info multipv 1 time 5741 nodes 44534886 nps 7757000 score cp 14 depth 20 pv e2e4
e7e5 b1c3 g8f6 g1f3 b8c6 f1c4 f6e4 c3e4 d7d5 c4d3 d5e4 d3e4 c6e7 e4d3 f7f6 e1g1
d8d5 d1e2 c8f5 d3c4 d5e4 e2e4 f5e4 d2d3 e4f3 g2f3
bestmove e2e4 ponder e7e5

5.IvanHoe-BetaWH_12_w64SSE4.2
info multipv 1 time 7192 nodes 56785058 nps 7895000 score cp 12 depth 20 pv e2e4
e7e5 g1f3 b8c6 f1c4 g8f6 b1c3 f8c5 e1g1 e8g8 d2d3 d7d6 c1e3 c5e3 f2e3 c8e6 c4e6
f7e6 d1d2 d8e7 c3b5 f6g4
info multipv 1 time 7192 nodes 56785058 nps 7895000 score cp 12 depth 20 pv e2e4
e7e5 g1f3 b8c6 f1c4 g8f6 b1c3 f8c5 e1g1 e8g8 d2d3 d7d6 c1e3 c5e3 f2e3 c8e6 c4e6
f7e6 d1d2 d8e7 c3b5 f6g4
bestmove e2e4 ponder e7e5

6.IvanHoe-BetaWH_12_w64SSSE3
info multipv 1 time 5600 nodes 43069889 nps 7691000 score cp 11 depth 20 pv e2e4
e7e5 g1f3 b8c6 f1c4 g8f6 d2d3 f8c5 e1g1 e8g8 b1c3 d7d6 c3a4 c8e6 a4c5 d6c5 c4e6
f7e6 c1e3 d8d6 c2c3 a8d8
info multipv 1 time 5600 nodes 43069889 nps 7691000 score cp 11 depth 20 pv e2e4
e7e5 g1f3 b8c6 f1c4 g8f6 d2d3 f8c5 e1g1 e8g8 b1c3 d7d6 c3a4 c8e6 a4c5 d6c5 c4e6
f7e6 c1e3 d8d6 c2c3 a8d8
bestmove e2e4 ponder e7e5

O2 Qx

1.IvanHoe-BetaWH_12_w64
info multipv 1 time 7987 nodes 62456131 nps 7819000 score cp 14 depth 20 pv d2d4
g8f6 b1c3 d7d5 c1f4 e7e6 e2e3 f8d6 g1f3 d6f4 e3f4 e8g8 f1d3 b8c6 e1g1 a7a6 d1d2
h7h6 h2h3 d8d6 a2a3 c8d7 a1e1
info multipv 1 time 7987 nodes 62456131 nps 7819000 score cp 14 depth 20 pv d2d4
g8f6 b1c3 d7d5 c1f4 e7e6 e2e3 f8d6 g1f3 d6f4 e3f4 e8g8 f1d3 b8c6 e1g1 a7a6 d1d2
h7h6 h2h3 d8d6 a2a3 c8d7 a1e1
bestmove d2d4 ponder g8f6

2.IvanHoe-BetaWH_12_w64SSE2
info multipv 1 time 8705 nodes 69706135 nps 8007000 score cp 14 depth 20 pv d2d4
d7d5 c1f4 e7e6 e2e3 g8f6 g1f3 f8d6 b1c3 d6f4 e3f4 e8g8 f1d3 b8c6 e1g1 d8d6 d1d2
a7a6 a2a3 h7h6 h2h3 c8d7 a1e1
info multipv 1 time 8705 nodes 69706135 nps 8007000 score cp 14 depth 20 pv d2d4
d7d5 c1f4 e7e6 e2e3 g8f6 g1f3 f8d6 b1c3 d6f4 e3f4 e8g8 f1d3 b8c6 e1g1 d8d6 d1d2
a7a6 a2a3 h7h6 h2h3 c8d7 a1e1
bestmove d2d4 ponder d7d5

3.IvanHoe-BetaWH_12_w64SSE3
info multipv 1 time 8424 nodes 67042599 nps 7958000 score cp 14 depth 20 pv d2d4
d7d5 b1c3 b8c6 c1f4 e7e6 e2e3 g8f6 g1f3 f8d6 f3e5 e8g8 f1d3 h7h6 e1g1 c8d7 c3b5
d8e7 b5d6 c7d6 e5c6 d7c6 a2a4
info multipv 1 time 8424 nodes 67042599 nps 7958000 score cp 14 depth 20 pv d2d4
d7d5 b1c3 b8c6 c1f4 e7e6 e2e3 g8f6 g1f3 f8d6 f3e5 e8g8 f1d3 h7h6 e1g1 c8d7 c3b5
d8e7 b5d6 c7d6 e5c6 d7c6 a2a4
bestmove d2d4 ponder d7d5

4.IvanHoe-BetaWH_12_w64SSE4.1
info multipv 1 time 4883 nodes 37923993 nps 7766000 score cp 11 depth 20 pv e2e4
e7e5 g1f3 b8c6 f1c4 f8c5 b1c3 g8f6 e1g1 e8g8 d2d3 d7d6 c3a4 c8e6 a4c5 d6c5 c4e6
f7e6 c1e3 d8d6 c2c3 a8d8
info multipv 1 time 4883 nodes 37923993 nps 7766000 score cp 11 depth 20 pv e2e4
e7e5 g1f3 b8c6 f1c4 f8c5 b1c3 g8f6 e1g1 e8g8 d2d3 d7d6 c3a4 c8e6 a4c5 d6c5 c4e6
f7e6 c1e3 d8d6 c2c3 a8d8
bestmove e2e4 ponder e7e5

5.IvanHoe-BetaWH_12_w64SSE4.2
info multipv 1 time 4821 nodes 36967060 nps 7667000 score cp 12 depth 20 pv e2e4
e7e5 g1f3 b8c6 f1c4 g8f6 d2d3 f8c5 b1c3 e8g8 c1e3 c5e3 f2e3 d7d6 e1g1 c8e6 c4e6
f7e6 d1d2 d8e7 c3b5 f6g4
info multipv 1 time 4821 nodes 36967060 nps 7667000 score cp 12 depth 20 pv e2e4
e7e5 g1f3 b8c6 f1c4 g8f6 d2d3 f8c5 b1c3 e8g8 c1e3 c5e3 f2e3 d7d6 e1g1 c8e6 c4e6
f7e6 d1d2 d8e7 c3b5 f6g4
bestmove e2e4 ponder e7e5

6.IvanHoe-BetaWH_12_w64SSSE3
info multipv 1 time 9111 nodes 73304125 nps 8045000 score cp 14 depth 20 pv d2d4
d7d5 g1f3 g8f6 c1f4 e7e6 e2e3 f8d6 b1c3 d6f4 e3f4 d8d6 f3e5 c8d7 d1d3 e8g8 e1c1
b8c6 c1b1 d6b4 a2a3 c6e5 f4e5
info multipv 1 time 9111 nodes 73304125 nps 8045000 score cp 14 depth 20 pv d2d4
d7d5 g1f3 g8f6 c1f4 e7e6 e2e3 f8d6 b1c3 d6f4 e3f4 d8d6 f3e5 c8d7 d1d3 e8g8 e1c1
b8c6 c1b1 d6b4 a2a3 c6e5 f4e5
bestmove d2d4 ponder d7d5

The last two SSE4.2 & SSSE3 i run again but now with "go depth 22" to confirm if SSSE3 is faster!

5.IvanHoe-BetaWH_12_w64SSE4.2
info time 27378 nodes 225393857 nps 8232000 hashfull 1000 cpuload 0
info multipv 1 time 28189 nodes 231815821 nps 8223000 score cp 13 depth 22 pv e2
e4 e7e5 g1f3 b8c6 b1c3 g8f6 f1e2 f8d6 e1g1 e8g8 d2d3 b7b6 h2h3 c8b7 c1e3 a7a6 d1
d2 h7h6 a2a3 d8e7 b2b4 e7e6
info multipv 1 time 28189 nodes 231815821 nps 8223000 score cp 13 depth 22 pv e2
e4 e7e5 g1f3 b8c6 b1c3 g8f6 f1e2 f8d6 e1g1 e8g8 d2d3 b7b6 h2h3 c8b7 c1e3 a7a6 d1
d2 h7h6 a2a3 d8e7 b2b4 e7e6
bestmove e2e4 ponder e7e5

6.IvanHoe-BetaWH_12_w64SSSE3
info multipv 1 time 36536 nodes 305786718 nps 8369000 score cp 15 depth 22 pv d2 <- Bill highest nodes/sec!!
d4 g8f6 c2c4 c7c5 e2e3 c5d4 e3d4 b8c6 g1f3 d7d5 b1c3 g7g6 c4d5 f6d5 d1b3 e7e6 f1
b5 f8g7 c1g5 d8b6 c3d5 e6d5 f3e5 e8g8 b5c6 b6b3 a2b3
info multipv 1 time 36536 nodes 305786718 nps 8369000 score cp 15 depth 22 pv d2 <- Bill highest nodes/sec!!
d4 g8f6 c2c4 c7c5 e2e3 c5d4 e3d4 b8c6 g1f3 d7d5 b1c3 g7g6 c4d5 f6d5 d1b3 e7e6 f1
b5 f8g7 c1g5 d8b6 c3d5 e6d5 f3e5 e8g8 b5c6 b6b3 a2b3
bestmove d2d4 ponder g8f6


compared to PeterPan's latest builds:

"go depth 22"

1.IvanHoe 53bU-x64 from Vlad0
info multipv 1 time 71636 nodes 587654367 nps 8203000 score cp 14 depth 22 pv d2
d4 g8f6 b1c3 d7d5 c1f4 e7e6 e2e3 f8d6 g1f3 d6f4 e3f4 e8g8 f1d3 d8d6 d1d2 a7a6 e1
g1 b8c6 a2a3
info multipv 1 time 71636 nodes 587654367 nps 8203000 score cp 14 depth 22 pv d2
d4 g8f6 b1c3 d7d5 c1f4 e7e6 e2e3 f8d6 g1f3 d6f4 e3f4 e8g8 f1d3 d8d6 d1d2 a7a6 e1
g1 b8c6 a2a3


2.IvanHoe-BPP0.1 from Peterpan!
info multipv 1 time 41138 nodes 376238168 nps 9145000:sm165: score cp 13 depth 22 pv d2
d4 g8f6 b1c3 d7d5 c1f4 e7e6 e2e3 f8d6 g1f3 d6f4 e3f4 e8g8 f1e2 b8c6 e1g1 c8d7 d1
d3 a7a6 a2a3 h7h6 f3e5 d8e7 h2h3 c6e5 f4e5
info multipv 1 time 41138 nodes 376238168 nps 9145000:sm165: score cp 13 depth 22 pv d2
d4 g8f6 b1c3 d7d5 c1f4 e7e6 e2e3 f8d6 g1f3 d6f4 e3f4 e8g8 f1e2 b8c6 e1g1 c8d7 d1
d3 a7a6 a2a3 h7h6 f3e5 d8e7 h2h3 c6e5 f4e5

PeterPan is already above 9Milion nodes/sec!


1) Take a set of code and find the best Intel switches and build.

2) Then take that same set of code and do a PGO with VTune
and get an engine that is 15% faster.

Do just a PGO and get about 10%.
Do just a VTune and get ??% {Is this PP's 11%?}

So, it looks like PP does a PGO or a VTune
but not both?, as i am guessing here.

Perhaps 9474550 nps would be a reasonable goal for this machine...

======================================== ==========

My 32-bit has SSSE3
and my 64-bit SSSE3 build
is fastest {engine speed}
on the test machine, curious.

However,
Intel needs a 64-bit machine
for max engine speed
since Intel compiler
integrates intensely with the machine
when it optimizes, while optimizing...
Intel Optimizations: PGO, VTune, plus many switch-settings possibilities...

halestorm
28.07.2010, 23:02
A update..still not loose yet!
Blitz 5m Core i7 920 @3.87Ghz , Blitz 5 0
Ivanhoe-B PP-0.1
Ipman.

Ipman

1. We have similar computer speeds. Do you have a page or guide for getting a 2.67 I7920 to 3.87 or 4.0? I'm at 3.5 stable.

2. Which build, exactly, of this engine are you using? There are four, I believe, and as I wrote earlier, I did not have good luck with Ivanhoe-B PP-0.1 on my I7.

Noobster
28.07.2010, 23:06
However,
Intel needs a 64-bit machine
for max engine speed
since Intel compiler
integrates intensely with the machine
when it optimizes, while optimizing...
Intel Optimizations: PGO, VTune, plus many switch-settings possibilities...

http://immortalchess.net/forum/showthread.php?t=2516
Some test by me, 30 minutes in start position.

Sandi
28.07.2010, 23:07
Skulltrail Intel QX 9775 Yorkfield (8X4090 MHz)
Fritz Benchmark: Relative speed = 38.78 . kN/s =18.476
Time 3'+0" .Hash 256 . Ponder off. Tablebases 5 pieces + Robbo Triple Bases
Book: HS-7moves.ctg.

IvanHoe T52A x64 - Deep Rybka 4 :
26-24 . 52% - 48%. ( +11 -9 =30)

The 50 games:

WHMoweryJr
28.07.2010, 23:12
Finally, I'm still noticing some problems when running "TitanicMode" and "benchmark". Some times, it remains freezed during one or 2 seconds, without any reason (and its final nodes per second is 300000, instead of 1 million). I'm pretty sure that this bug is related to the problem that Vlad0 spent so much time fixing, and probably is the cause of the losses in time when blitzing.

1) Timing: Actually the CheckDone()
in make_move.c
needs to always occur.

That eliminates alotta timeouts in my versions {over 99% of them}.

2) Timing: In alt timing in my versions i test for
stuck-in-first-root-move,
and if time is too far gone,
search ends. AlternateTiming is default ON.

These things in my versions
have solved the timing problems.

3) Sleep-problem: I have tried limiting the WAIT_CON_LOCK time down to 1 second
from INFINITE, perhaps that will solve the problem,
as 1 second is fast relative to our games
but slow relative to clock rate,
1sec is over 3,000,000,000 clock cycles
on most machines these days.


Glossary:
Timing == cause of the losses in time when blitzing
Sleep-problem == it remains freezed during one or 2 seconds, without any reason

TCH
28.07.2010, 23:15
The correlation coefficient between ELO and go depth 22 speed will be much closer to 1 than 0.5 and way above 0.
I had a paradoxical example - 5 of your compilations under the IC and 3 in the VC, the rate divided into two groups (Core2Duo): All IC approximately 670-690 kN / s, all the VC about 735-745 kN / s. I'm playing a tournament for 20 seconds / play with them without the bases (10 rounds). Result: all 3 fast VC gaining 30-35 points, all 5 IC 50-55 points. That is, two dense groups, but the results are opposite to the expected speed. How do you explain - I do not know.

Ipman
28.07.2010, 23:32
Ipman

1. We have similar computer speeds. Do you have a page or guide for getting a 2.67 I7920 to 3.87 or 4.0? I'm at 3.5 stable.

2. Which build, exactly, of this engine are you using? There are four, I believe, and as I wrote earlier, I did not have good luck with Ivanhoe-B PP-0.1 on my I7.

Hi halestorm,

1.I have learned everything about overclocking on Xtremesystems..your overclock depends on all your parts in your computer.
A big help is..make that you know your Batch number from your cpu..type this number in google search and you will get information from other people who has the same cpu and their overclock!

2.If you look to the message 3118 on page 156 you can read that PP has 3 versions with different compiles and that the first one is the 4.2 who is usefull for our i7!

Ipman.

WHMoweryJr
28.07.2010, 23:41
I had a paradoxical example - 5 of your compilations under the IC and 3 in the VC, the rate divided into two groups (Core2Duo): All IC approximately 670-690 kN / s, all the VC about 735-745 kN / s. I'm playing a tournament for 20 seconds / play with them without the bases (10 rounds). Result: all 3 fast VC gaining 30-35 points, all 5 IC 50-55 points. That is, two dense groups, but the results are opposite to the expected speed. How do you explain - I do not know.

The faster Intel builds will outscore the slower Intel builds.
The faster Microsoft builds will outscore the slower Microsoft builds.

Please post the nps at startpos on your machine
for all of the entrants
perhaps after 60 seconds just grab the peak
insuring your machine is like fresh-snow {unoccupied}
and insure the results repeatable,
ie get the same peak at least a few times
to know that the number is good.

The faster Intel builds will outscore the slower Intel builds.
The faster Microsoft builds will outscore the slower Microsoft builds.

15sec games will timeout, but 15sec games with 1/2 sec increment or more
will not. Timeouts not very good for this experiment, so tourney should be setup without timeouts...

The faster Intel builds will outscore the slower Intel builds.
The faster Microsoft builds will outscore the slower Microsoft builds.

Intel is a "more solid" compiler than Microsoft {on Intel machines}
since Intel can develop on their own machines
AND
all the necessary literature/documentation
that the compiler builders/designers/coders need
is right at Intel's fingertips,
Microsoft can only outsource
and must guess when Intel's documentation
on the Intel machines is lacking,
and have you seen Intel's documentation lately?
So, Microsoft would have to have a team
in-house at Intel to develop a compiler
as solid as Intel's! Not gonna happen.

The faster Intel builds will outscore the slower Intel builds.
The faster Microsoft builds will outscore the slower Microsoft builds.

After many hundreds of games,
the faster ones will score higher,
std dev == sqrt(2 * N) for large N
{we are all conclusionalizing from far too few games}

The faster Intel builds will outscore the slower Intel builds.
The faster Microsoft builds will outscore the slower Microsoft builds.

And my Intel builds were built with a Profile Guided Optimization {PGO}
where data from actual games was used by the Intel compiler.
So, in this case, assuming people would all use the Intel compiler
at least all 32-bit Intel machines, i did not do so great a job with the Microsoft compile,
just using startpos - which may explain why startpos is faster???

The Intel PGO builds should all search more nodes across a game for the given time searched
across an ensemble of games,
ie. across an ensemble of games, should achieve a higher average nps . . .

Vlad0
28.07.2010, 23:42
Finally, I'm still noticing some problems when running "TitanicMode" and "benchmark". Some times, it remains freezed during one or 2 seconds, without any reason (and its final nodes per second is 300000, instead of 1 million).


It seems to me that You just forgot to reset the number of Threads from
the default 1 to the number of cores on Your PC. Just setting
the "Titanic mode" is not enough :)


I'm pretty sure that this bug is related to the problem that Vlad0 spent so much time fixing, and probably is the cause of the losses in time when blitzing.

This bug (falling of the engine in the eternal asleep) is already corrected in oficial 999952 beta version, so nothing needs to be done to fix it.

slankamen
28.07.2010, 23:57
Intel(R) Core(TM) i7 4CPU 960 3.20GHz8x @ 4.005 GHz with 4,096 MB Memory
Microsoft Windows XP 64 Bit Professional Service Pack 2 (Build 3790)
Fritz Benchmark:
Speed: 24.98
KNS: 11988
GUI: CB Rybka 3
Book: Perfect 2009 10 moves
Hash: 256
RB and TB: ON
Ponder: OFF

Blitz:10' 0


1 Deep Rybka 4 SSE42 x64 3190 +6/-2/=12 60.00% 12.0/20
2 IvanHoe T52A.x64 SSE4.2 3195 +2/-6/=12 40.00% 8.0/20

jc.m
29.07.2010, 00:03
It seems to me that You just forgot to reset the number of Threads from
the default 1 to the number of cores on Your PC. Just setting
the "Titanic mode" is not enough :)


When I do my tests, I run:
setoption name TitanicMode value true
setoption name TitanicCPUsSizings value 2
benchmark
(2 because I have a dual-core)

One or 2 times, I noticed that the test freezed during a few seconds, giving the terrible averave nodes count of 300000 per second (1/3 of the standard value).

Also, another question I have is: why do I have to disable aliasing (/Oa) with Intel C++ 11.1 to get stable releases ?
Vlad0, do you have similar problems with Intel C++ 12 ?

About the bug: yes, I read the thread, and verified that it was fixed in the release, but we still have concurrency problems, with games lost due to time.

Vlad0
29.07.2010, 00:13
I've just added as attachment to my esterday's blog my compilation IVH 52bU as well
(made this night). From the standard 200 games blitz 1+0 it follows, that this
compilation seeems to be worse than IVH 53bU:

1 Ivanhoe 52bU-x64 +11 +38/=125/-32 51.54% 100.5/195
2 IvanHoe 9.53b x64 -11 +32/=125/-38 48.46% 94.5/195

(with 3 "no move" bugs caused by 52bU and 2 -- by yesterdays KLO's 9.53b),
but remains yet slightly better than 9.53b.

I am asking somebody with admin rights: upload plz both these versions to some accessible public download sitе.

I am wondering, why the admins of this forum,
permitting the files to be attached to blogs (which is very convinient!), gave no
permission for all registered members of the forum to download these attachments!

PS I forgot to mention that on both IVH 53bU and IVH 52bU both "benchmark" and "smptest"
are running without any problems. All autodetection functions (number of threads, popcnt
and Robbobases) are present as well.

Добавлено через 8 минут
Also, another question I have is: why do I have to disable aliasing (/Oa) with Intel C++ 11.1 to get stable releases ?
Vlad0, do you have similar problems with Intel C++ 12 ?


No. But I was not changing any of the default ICC optimisation options.

Sandi
29.07.2010, 00:53
Intel X5355 Octa (8X2660 MHz)
Time 3'+0". Hash 256. GUI Fritz 12.Tablebases 5 pieces + Robbo Triple Bases.
Book: HS-7moves.ctg.

IvanHoe 9.52b x64 - Deep Rybka 4 :
25.5 - 24.5 . 51% - 49% (+11 -10 =29)

The 50 games:

immortal223
29.07.2010, 00:53
I am wondering, why the admins of this forum,
permitting the files to be attached to blogs (which is very convinient!), gave no
permission for all registered members of the forum to download these attachments!

done! It was my fault, I missed this setting

slankamen
29.07.2010, 01:20
Intel(R) Core(TM)2 Quad Q9550 2.83GHz4x @3500 MHz 4,096 MB Memory
Microsoft Windows XP 64 Bit Professional Service Pack 2 (Build 3790)
Fritz Benchmark:
Speed: 20.90
KNS: 10032
GUI: CB Rybka 3
Book: Perfect 2009 10 moves
Hash: 256
RB and TB: ON
Ponder: OFF

Blitz:15' 0


1 Deep Rybka 4 x64 3190 +6/-2/=12 60.00% 12.0/20
2 IvanHoe-Beta version 999952a- 3185 +2/-6/=12 40.00% 8.0/20

Rendruk
29.07.2010, 01:22
Intel(R) Core(TM)2 Quad Q9550 2.83GHz4x @3500 MHz 4,096 MB Memory
Microsoft Windows XP 64 Bit Professional Service Pack 2 (Build 3790)
Fritz Benchmark:
Speed: 20.90
KNS: 10032
GUI: CB Rybka 3
Book: Perfect 2009 10 moves
Hash: 256
RB and TB: ON
Ponder: OFF

Blitz:15' 0


1 Deep Rybka 4 x64 3190 +6/-2/=12 60.00% 12.0/20
2 IvanHoe-Beta version 999952a- 3185 +2/-6/=12 40.00% 8.0/20

I was just thinking of picking a 53 version, and 52 comes out. Now I have to wait to see which compiles are good, if any, all over again haha!

TCH
29.07.2010, 01:23
IvanHoe52bU(I) from Vlad0
http://multi-up.com/311992

TCH
29.07.2010, 02:11
Vlad0, в IvanHoe52bU(I) роббобазы не грузятся (в 53-й работали)

slankamen
29.07.2010, 04:35
AMD Phenom(tm) II X6 1090T Processor 6x 3.895 GHz 4,096 MB Memory
Microsoft Windows XP 64 Bit Professional Service Pack 2 (Build 3790)
Fritz Benchmark:
Speed: 28.33
KNS: 13599
GUI: CB Rybka 3
Book: Perfect 2009 10 moves
Hash: 256
RB and TB: ON
Ponder: OFF

Blitz:10' 0


1 Deep Rybka 4 x64 3185 +3/-1/=16 55.00% 11.0/20
2 Ivanhoe-B PP-0.1POPCNT 3195 +1/-3/=16 45.00% 9.0/20

BTO7
29.07.2010, 04:43
Hi and thanks for the feedback !

I'll try to answer to everybody in a single answer, please take time to read the answers, they might be interesting for the future...


This is not a direction I'd like to take.
If compilers have tricks to improve the engine, it should be added directly in Ivanhoe's original version, not in the Windows' compiles.
First, having a Windows version stronger than the Linux one is not fair for the original version, which was created to share a strong open-source engine.

Jc.....something about this answer bothers me a bit. While I understand your wanting to be fair to the decembrists and as you know getting along with their door lady at the site is not very welcoming. Second your suggesting NOT to allow windows version to surpass linux? So what then is everyone doing here? Linux version should not wait on windows nor should windows wait on Linux IMHO. They share the source freely and there is nothing that says ...you can only use this source if you do not modify it beyond our Linux versions. The idea in my mind it to make the source the best possible period. I don't understand the fair your talking about in this sense. If the rules are use our code but dont surpass our linux compiles everyone is wasting their time. If the windows version by a team winds up stronger then the Linux version in my book of open source ....thats just the way it is. Angels over there have this very attitude of its free BUT dont do this or that with it. Either one or a group aims to make it the best they can or there is no use in bothering. Its a free open source to do as compiliers wish and nobody wants the restraint of holding back to not make it better then the Linux version....I'm sorry but i dont agree with that one bit. They release the code in hopes of it being taken to the next level in some fashion...thats why its open source to begin with no?

Regards
BT

Rendruk
29.07.2010, 05:08
Jc.....something about this answer bothers me a bit. While I understand your wanting to be fair to the decembrists and as you know getting along with their door lady at the site is not very welcoming. Second your suggesting NOT to allow windows version to surpass linux? So what then is everyone doing here? Linux version should not wait on windows nor should windows wait on Linux IMHO. They share the source freely and there is nothing that says ...you can only use this source if you do not modify it beyond our Linux versions. The idea in my mind it to make the source the best possible period. I don't understand the fair your talking about in this sense. If the rules are use our code but dont surpass our linux compiles everyone is wasting their time. If the windows version by a team winds up stronger then the Linux version in my book of open source ....thats just the way it is. Angels over there have this very attitude of its free BUT dont do this or that with it. Either one or a group aims to make it the best they can or there is no use in bothering. Its a free open source to do as compiliers wish and nobody wants the restraint of holding back to not make it better then the Linux version....I'm sorry but i dont agree with that one bit. They release the code in hopes of it being taken to the next level in some fashion...thats why its open source to begin with no?

Regards
BT

Seriously, why would someone not use Windows?

If there's something you don't like about it you can always customize your own Windows version with this free tool from the Rockerz website.

http://www.rtwincustomize.net/web/viewforum.php?f=110&start=0

Pretty great application. You can customize it so the file size gets down to 700 megabytes if you wanted to take enough stuff out, plus it lets you add programs to be installed with windows, like say you want it to come with firefox and arena and who knows what else pre-installed, or even all the windows updates so you don't have to go through that long wait later on. . . I could give a list of features but if you're interested in that sort of thing just go check it out. It's good sauce.

jeandis
29.07.2010, 05:47
IvanHoe52bU(I) from Vlad0
http://multi-up.com/311992

this engine is weaker than Ivanhoe B PP.0.1

Vega Odinov
29.07.2010, 06:24
http://www.speedyshare.com/files/235...0.1Intel64.rar
http://www.speedyshare.com/files/235...PP0.1AMD64.rar
http://www.speedyshare.com/files/235...P0.132bits.rar

this is the old ivanhoe BPP.0.1 links
URL NOT found .. can someone please upload the intel 64 + 32 bits ?
thanks

Vega Odinov
29.07.2010, 06:28
Sadly .. IvanHoe 52bU I is a little weaker than IvanHoe 53bU I and 57aU fixed version

one more thing .. can someone upload IvanHoe 63 Mod5-Turbo-x64-i7 ?
thanks

slankamen
29.07.2010, 06:35
Intel(R) Core(TM) i7 4CPU 960 3.20GHz8x @ 4.005 GHz with 4,096 MB Memory
Microsoft Windows XP 64 Bit Professional Service Pack 2 (Build 3790)
Fritz Benchmark:
Speed: 24.98
KNS: 11988
GUI: CB Rybka 3
Book: Perfect 2009 10 moves
Hash: 256
RB and TB: ON
Ponder: OFF

Blitz:10' 0


1 Deep Rybka 4 SSE42 x64 3190 +2/-1/=17 52.50% 10.5/20
2 Ivanhoe-B PP-0.1.x64 3195 +1/-2/=17 47.50% 9.5/20

slankamen
29.07.2010, 07:56
Intel(R) Core(TM)2 Quad Q9550 2.83GHz4x @3500 MHz 4,096 MB Memory
Microsoft Windows XP 64 Bit Professional Service Pack 2 (Build 3790)
Fritz Benchmark:
Speed: 20.90
KNS: 10032
GUI: CB Rybka 3
Book: Perfect 2009 10 moves
Hash: 256
RB and TB: ON
Ponder: OFF
Blitz:10' 0


1 IvanHoe 9.52b x64 KLO 3185 +5/-4/=11 52.50% 10.5/20
2 Deep Rybka 4 x64 3190 +4/-5/=11 47.50% 9.5/20

yayo
29.07.2010, 09:04
http://www.speedyshare.com/files/235...0.1Intel64.rar
http://www.speedyshare.com/files/235...PP0.1AMD64.rar
http://www.speedyshare.com/files/235...P0.132bits.rar

this is the old ivanhoe BPP.0.1 links
URL NOT found .. can someone please upload the intel 64 + 32 bits ?
thanks

not sure if this is it van. h?

Ipman
29.07.2010, 12:47
done! It was my fault, I missed this setting

Thanks immortal223..now we can finally download from Vlad0 Blog site!!

http://immortalchess.net/forum/blog.php?u=971

Ipman.

calu
29.07.2010, 13:25
please try to test ivanhoe Bpp0.1 if you can on 15΄+ 10" blitz... i think that the engine performs even better in higher time controls !

Ipman
29.07.2010, 13:30
Look if you start the .exe from Vlad0 ,just perfect!!

a little detail..is it not possible to see the overclocked speed from the system!

Ivanhoe 52bU-x64 (by Vlad0)
OS: Windows 7 Ultimate
CPU: Intel(R) Core(TM) i7 CPU 920 @ 2.67GHz
Cores: 4, MaxThreads: 4
CPU has Popcnt :)
Triple bases dir: E:\RobboBases\RobboTripleBase
Total bases dir: E:\RobboBases\RobboTotalBase
TotalBaseCache is 1mb + (1mb)

Just great,i like this very much..

Ipman.

jc.m
29.07.2010, 14:38
Jc.....something about this answer bothers me a bit.

You are thinking too much...
People who think too much don't act enough.
I like to mention the chain: thought/words/action.
As long as you remain in the thought or words stages, nothing is done.

Angels over there have this very attitude of its free BUT dont do this or that with it

Angels are the stupidest trolls and I won't ever go back to Ippolit's site due to their shitty behavior.
They are stupid enough to not realize that they use Windows, and they are unable to make their own compiles.

Second your suggesting NOT to allow windows version to surpass linux?

Wow, where did you read this ?
I'll put more details just below.

Linux version should not wait on windows nor should windows wait on Linux

Sure, but as a long time coder, I'd like to share some of my past experience.
15 years ago, I was the lead programmer of a game, and at this time, there was no SVN nor Visual SourceSafe.
We were 6 coders, and everybody was working on his local sources.
I was in charge to manually merge all the source code daily to have the latest release.
And believe me, it was very time-consuming.
Now, serious developers use tools like SVN or Git or Mercurial, but merging branches of different sources always takes a lot of time.
So if you really want, I can easily create a branch for your version, but you'll have to maintain it yourself, and in no case it will be merged into the trunk.

This means that you can make all your stable or unstable releases on your own, but all your changes may disappear after a new version of Ivanhoe's.

The purpose of create a source repository was to allow the creation of a team, but I see that nobody really cares to work in common, and add new functionalities to the trunk.
As a long-time developer (I started working 26 years ago), I know that if there is no team, people will just come in and out without any direction.
How many people will pass and quit before we have a team here ?

Also, I do not see a lot of code being shared here.
For example, I publish my changes, and WMHowery does also, but we are not the majority of compilers.
As long as the code is not shared, there won't be an incentive to make Ivanhoe better.
Look at the clones of Ivanhoe: Firebird and Houdini. The coders behind them are perhaps great (though from my point of view, they have serious ego problems), but they are alone.

Frankly, outside of Vlad0, nobody tested the SVN I created, so what will happen as long as people continue working on their own version, without sharing with others ?
The answer is: nothing !

Seriously, why would someone not use Windows?

If there's something you don't like about it you can always customize your own Windows version with this free tool from the Rockerz website.


Frankly, and I speak as a Windows developer, Windows is a PITA (I let you search for this acronym).
I installed my laptop with Windows 64 this week-end, and I realize now that I need to do a lot of deep checks to improve its performance.
For example, when my computer is idling, the fan tends to start and stop at some intervals, even though I have no running services (fans are a clue about peaks of computing power).
I found a (very technical) page explaining how to locate the problem:
http://blogs.technet.com/b/markrussinovich/archive/2008/04/07/3031251.aspx
Why doesn't it simply work ? Why do I have to track the incorrect driver manually ?

I think Ubuntu has a lot less problems (and yes, I'm using it in my virtual machines).

Vlad0
29.07.2010, 15:36
done! It was my fault, I missed this setting

Thank You very much! Now everyone can place all his compiles in one place,
as an attachment to his blog. And all engines can be very easily found --
one needs no more to search the download site in dozens of IVH pages :).
And no need to download engines on these sites as well :) Great!

BTO7
29.07.2010, 15:54
You are thinking too much...
People who think too much don't act enough.
I like to mention the chain: thought/words/action.
As long as you remain in the thought or words stages, nothing is done.



Angels are the stupidest trolls and I won't ever go back to Ippolit's site due to their shitty behavior.
They are stupid enough to not realize that they use Windows, and they are unable to make they own compiles.



Wow, where did you read this ?
I'll put more details just below.



Sure, but as a long time coder, I'd like to share some of my past experience.
15 years ago, I was the lead programmer of a game, and at this time, there was no SVN nor Visual SourceSafe.
We were 6 coders, and everybody was working on his local sources.
I was in charge to manually merge all the source code daily to have the latest release.
And believe me, it was very time-consuming.
Now, serious developers use tools like SVN or Git or Mercurial, but merging branches of different sources always takes a lot of time.
So if you really want, I can easily create a branch for your version, but you'll have to maintain it yourself, and in no case it will be merged into the trunk.

This means that you can make all your stable or unstable releases on your own, but all your changes may disappear after a new version of Ivanhoe's.

The purpose of create a source repository was to allow the creation of a team, but I see that nobody really cares to work in common, and add new functionalities to the trunk.
As a long-time developer (I started working 26 years ago), I know that if there is no team, people will just come in and out without any direction.
How many people will pass and quit before we have a team here ?

Also, I do not see a lot of code being shared here.
For example, I publish my changes, and WMHowery does also, but we are not the majority of compilers.
As long as the code is not shared, there won't be an incentive to make Ivanhoe better.
Look at the clones of Ivanhoe: Firebird and Houdini. The coders behind them are perhaps great (though from my point of view, they have serious ego problems), but they are alone.

Frankly, outside of Vlad0, nobody tested the SVN I created, so what will happen as long as people continue working on their own version, without sharing with others ?
The answer is: nothing !



Frankly, and I speak as a Windows developer, Windows is a PITA (I let you search for this acronym).
I installed my laptop with Windows 64 this week-end, and I realize now that I need to do a lot of deep checks to improve its performance.
For example, when my computer is idling, the fan tends to start and stop at some intervals, even though I have no running services (fans are a clue about peaks of computing power).
I found a (very technical) page explaining how to locate the problem:
http://blogs.technet.com/b/markrussinovich/archive/2008/04/07/3031251.aspx
Why doesn't it simply work ? Why do I have to track the incorrect driver manually ?

I think Ubuntu has a lot less problems (and yes, I'm using it in my virtual machines).

I'm sorry jc ... I must of misunderstood you. The trunk idea is what i was all for....meaning the team develops from the trunk and i would be more interested in testing compiles of a trunk based team effort then a bunch of compiles basically the same from different people. What you and WHM are doing is just the thing im talking about. Working together and coming up with the best Ivan so we as testers can focus on just a few compiles. So many releases so fast and too many people building them causes some chaos and i dont want to offend anyone by not testing their compiles but where to start and with whos compile. Testing takes lots of computer time ...so all I was trying to say is I would much rather test a version by say you WHM PP VladO and company from a team effort and a single version by all of you. Then we could really test the hell out if it. A new source a week and 7 compilers all releasing practically together ...there is hardly time to figure out which one is best before new code is out. I thought you had said you didnt like to go in that direction so to me I thought you were trying to say ...just let each guy do his own thing and report back to decembrists. It would be nice to have a good stable compile.....take Ahmeds tuning your tricks Vlado's tricks WHM's speed and the stability of PP or JR's compiles and i would think we would have the best compile overall....and then just one killer version to test. All info going back to decembrists of course. Its a shame this cant be done. To me strength is not improving much for awhile ....so if there is a tiny bit chance to gain anything i would be a team effort compile I guess is all im saying. I personally like to test for strength, speed and stability. Unless everyones just compiling to find bugs only ...in that case I would like to wait a bit until something solid arrives is all.

TC
BT

slankamen
29.07.2010, 16:46
Intel(R) Core(TM)2 Quad Q9550 2.83GHz4x @3500 MHz 4,096 MB Memory
Microsoft Windows XP 64 Bit Professional Service Pack 2 (Build 3790)
Fritz Benchmark:
Speed: 20.90
KNS: 10032
GUI: CB Rybka 3
Book: Perfect 2009 10 moves
Hash: 256
RB and TB: ON
Ponder: OFF
Blitz:10' 0


1 IvanHoe 9.52b x64 KLO 3185 +6/-4/=20 53.33% 16.0/30
2 Deep Rybka 4 x64 3190 +4/-6/=20 46.67% 14.0/30

Добавлено через 2 минуты
Score from 50 games


1 IvanHoe 9.52b x64 KLO 3185 +11/-8/=31 53.00% 26.5/50
2 Deep Rybka 4 x64 3190 +8/-11/=31 47.00% 23.5/50

Sandi
29.07.2010, 21:11
THE CHAMPION OF BLITZ 8 THREADS AGAINST THE CAPITALIST :

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Skulltrail Intel QX 9775 (8X4090 MHz),
Fritz Benchmark: Relative speed = 38,87. K/n/s = 18.476.
Time 20' (equivalent to 50' on a standard quad).

Hash 1024. GUI Fritz 12. Ponder off. Tablebases 3-4-5 + Robbo Triple Bases
Book : IPPOLIT 7-moves.ctg.

IvanHoe T 55A (no SSE) x64 - Deep Rybka 4 :
17 -13 . ( +9 - 5 = 16 )
Very good result also in long time control !:sm75:

Rendruk
29.07.2010, 21:20
Frankly, and I speak as a Windows developer, Windows is a PITA (I let you search for this acronym).
I installed my laptop with Windows 64 this week-end, and I realize now that I need to do a lot of deep checks to improve its performance.
For example, when my computer is idling, the fan tends to start and stop at some intervals, even though I have no running services (fans are a clue about peaks of computing power).
I found a (very technical) page explaining how to locate the problem:
http://blogs.technet.com/b/markrussi...7/3031251.aspx
Why doesn't it simply work ? Why do I have to track the incorrect driver manually ?

I think Ubuntu has a lot less problems (and yes, I'm using it in my virtual machines).


Hey, the problem, or at least a solution, is in your Power settings. Try and get them optimized for laptops. Set it to hibernate or go to sleep.

Go to control panel, power settings (or power options), advanced settings (or "change plan settings", (it all depends on your windows version)

Once there also be sure to doublecheck your processor settings, in windows 7 it is named "Processor power management"

Your problem might be located there. Windows does adjust fan speed based upon cpu idle time. (But the other settings will help more I'm sure)

Also, if i'm completely wrong, double-check your bios settings. Some of them might work differently with a new OS installed.

Never had a problem with my fans and windows though, so don't blame it on windows;P Even if it does end up being windows fault it would be the "Exception to the Norm", of course, in human psychology it is noted that first impressions are almost impossible to turn around. For example, if you meet a person and they make a bad first impression on you, then the next time you meet them, they make a good impression, the new data gathered on the person will not likely influence your overall judgement of them. While conversely, if you meet someone for the first time and they make a *good* impression, and the second time they do something kinnda fucked up, you'll give them the benefit of the doubt anyway. Probably works the same way with Windows:P

'Cuz peoples minds don't want to experience the feeling of being wrong about anything, so they tend to be more prone in rejecting information that conflicts with pre-existing beliefs. Like, you want to believe all the time you spent in Ubuntu wasn't wasted, so now you'll be more prone to find flaws in other OS's.

jc.m
29.07.2010, 21:31
Hey, the problem is in your Power settings.

Go to control panel, power settings (or power options), advanced settings (or "change plan settings", (it all depends on your windows version)


Thanks, but I discovered the problem.
Somebody had the same problem as me:
http://forums.pcper.com/showthread.php?t=466616

The problem appears when you have 2 network cards (I have a Wifi and an Ethernet cards), and it's an old Windows bug, which was here since Windows 2003.
The trick is to disable LMHOSTS (by changing a registry value), otherwise Windows tries to read the file lmhosts around 1 million times every 10 minutes. The only problem is that this file doesn't exist on Windows 7 !!!

About the CPU fan, my PC is the HP serie where the NVidia videocard burns everything when running intensive 3D games, so they forced the fans to run at top speed whenever there is a little heat.


... you'll give them the benefit of the doubt anyway. Probably works the same way with Windows:P

Yes, it's a well known fact, but this works for people, not for software.
I always liked Windows XP and always hated Vista (and at my job, I was the only one forced to migrate, since nobody wanted to test it).

Edit: if you use Windows 7, I recommend that you update your drivers.
Here is a nice free site for that: http://www.ma-config.com/en/home/
I recommend running it with IE, and it installs a small ActiveX that checks your drivers and gives you links for the latest ones.

jc.m
29.07.2010, 21:43
Thank You very much! Now everyone can place all his compiles in one place,
as an attachment to his blog. And all engines can be very easily found --
one needs no more to search the download site in dozens of IVH pages :).
And no need to download engines on these sites as well :) Great!

I think this is not a very good idea, since this site tends to vanish regularly.
Today, during 3 hours, I got 502 errors from the front page, and I get errors like this almost every day.

BTW, did you hear about the Tactico engine ? (link: http://www.mediafire.com/?akkcvjd7bgreyz8)
It's a rip-off of an old version of Ivanhoe, and it's closed source.
The author forgot to remove the strings:

id author Guiyo TacticoLG LG3350Elo
id copyright Guiyo TacticoLG LG3350Elo, 92th year from Revolution, PUBLICDOMAIN (workers)
id dedicatory To Vladimir Ilyich

from the binary !!!
And of course, no mention about Ivanhoe or Ippolit.

Vlad0
29.07.2010, 21:51
There is distributed also "Deep Tactico 2010 Pro X64.exe" -- obtained by changing
some strings in Houdini 1.03 by Hex editor :) So there is no "closed source" in this case at all:
just a cheap fake. I suppose Tactico is produced by the same elementary trick.
May be by appending additionally some binary file to the end of hex edited executable
in order to change its length as well.

jc.m
29.07.2010, 22:17
There is distributed also "Deep Tactico 2010 Pro X64.exe" -- obtained by changing
some strings in Houdini 1.03 by Hex editor :) So there is no "closed source" in this case at all:
just a cheap fake. I suppose Tactico is produced by the same elementary trick.


Woah, you are right !
I didn't think it was possible to be this lame, but the author binary edited the files. His name is "Luis Olmos" (note the 3 spaces between his first and last names :D).
Also, the compilation path is: R:\Schaak\Tactico\VS\4MP\Release\Tactico _64.pdb
And Schaak means chess in dutch.

Rendruk
29.07.2010, 22:33
Only an idiot would buy his engine anyway, as it will never do better than a free one. . .

Which makes me wonder why people still buy anything from Rybka.

slankamen
29.07.2010, 23:00
AMD Phenom(tm) II X6 1090T Processor6x @ 3.200 GHz 4,096 MB Memory
Windows 7 Home Premium Home Edition (Build 7600)
Fritz Benchmark:
Speed: 23.33
KNS: 11196
GUI: CB Rybka 3
Book: Perfect 2009 10 moves
Hash: 256
RB and TB: ON
Ponder: OFF

Blitz:10' 0

123456789012345678
1 Ivanhoe 52bU-x64 3195 ½½½½½½11½½½½½½10½½ 10.0/18
2 Deep Rybka 4 x64 3185 ½½½½½½00½½½½½½01½½ 8.0/18

slankamen
29.07.2010, 23:19
Intel(R) Core(TM) i7 4CPU 960 3.20GHz8x @ 4.005 GHz with 4,096 MB Memory
Microsoft Windows XP 64 Bit Professional Service Pack 2 (Build 3790)
Fritz Benchmark:
Speed: 24.98
KNS: 11988
GUI: CB Rybka 3
Book: Perfect 2009 10 moves
Hash: 256
RB and TB: ON
Ponder: OFF

Blitz:10' 0

1 IvanHoe 9.52b x64KLO POPCNT 3195 +3/-2/=15 52.50% 10.5/20
2 Deep Rybka 4 SSE42 x64 3190 +2/-3/=15 47.50% 9.5/20

slankamen
29.07.2010, 23:51
[Event "AMD Phenom II 1090T.x64 , Blitz:10'"]
[Site "AMD Phenom II x6 1090T.x64 3."]
[Date "2010.07.29"]
[Round "13"]
[White "IvanHoe 9.52b x64KLO"]
[Black "Deep Rybka 4 x64"]
[Result "0-1"]
[ECO "C49"]
[WhiteElo "3195"]
[BlackElo "3185"]
[Annotator "0.18;0.20"]
[PlyCount "220"]
[TimeControl "600"]

{AMD Phenom(tm) II X6 1090T Processor 3200 MHz W=22.1 ply; 8,842kN/s; 1,625,
389 TBAs; Perfect 2009.ctg B=17.3 ply; 374kN/s; 436 TBAs; Perfect 2009.ctg} 1.
e4 {B/0 0} e5 {B/0 0} 2. Nf3 {B/0 0} Nc6 {B/0 0} 3. Bb5 {B/0 0} Nf6 {B/0 0} 4.
Nc3 {B/0 0} Bb4 {B/0 0} 5. O-O {B/0 0} O-O {B/0 0} 6. d3 {B/0 0} d6 {B/0 0} 7.
Bg5 {B/0 0} Bxc3 {B/0 0} 8. bxc3 {B/0 0} Qe7 {B/0 0} 9. Re1 {B/0 0} Nd8 {B/0 0}
10. d4 {B/0 0} Ne6 {Both last book move B/0 0} 11. Bd2 {0.18/21 26} c6 {0.20/
17 18} 12. Bd3 {0.24/20 9} Qc7 {0.20/17 20} 13. Qc1 {0.16/21 55} Re8 {0.16/17
22} 14. Qb2 {0.19/20 5} h6 {0.16/17 9} 15. h3 {0.24/22 47} b6 {0.22/18 47} 16.
a4 {0.24/21 16} Rb8 {0.17/17 17} 17. Rab1 {0.20/21 52} Bd7 {(a7-a5) 0.13/16 22}
18. Qa1 {(Qb2-a3) 0.18/21 24} Nh5 {(Ne6-f8) 0.08/17 32} 19. c4 {(Bd3-f1) 0.22/
19 3} Nef4 {(e5xd4) 0.13/17 29} 20. dxe5 {(Bd3-f1) 0.33/20 11} Bxh3 {(d6xe5) 0.
00/14 8} 21. Bxf4 {(g2xh3) 0.59/21 14} Nxf4 {0.27/14 6} 22. gxh3 {0.57/22 16}
Nxh3+ {0.36/14 5} 23. Kf1 {(Kg1-h2) 0.60/22 4} dxe5 {0.10/17 9} 24. Re3 {0.61/
21 3} Nf4 {0.08/16 4} 25. Nh4 {0.58/20 3} g6 {0.12/15 11} 26. Rg3 {0.57/19 5}
h5 {(Kg8-h7) 0.15/15 13} 27. Qc3 {(Nh4-g2) 0.57/20 4} Kh7 {0.22/16 40} 28. Ng2
{(Qc3-a3) 0.53/20 50} Ne6 {0.14/16 14} 29. Ne1 {0.54/19 8} Nc5 {0.19/15 6} 30.
Ra1 {0.47/20 15} a5 {0.19/16 19} 31. Nf3 {(Kf1-g2) 0.55/20 13} Qe7 {0.20/16 16}
32. Kg2 {0.55/20 9} Qf6 {0.20/15 9} 33. Qd2 {(Qc3-a3) 0.53/20 18} Kg7 {0.20/16
10} 34. Rh1 {0.60/19 3} Ne6 {(Re8-h8) 0.20/16 7} 35. Kf1 {(Kg2-g1) 0.60/20 7}
Nf4 {0.21/16 5} 36. Nh4 {(Qd2-e3) 0.61/20 3} Kf8 {(Re8-d8) 0.23/16 10} 37. Rf3
{(Nh4-g2) 0.61/21 10} Qg5 {0.21/16 6} 38. Qc1 {0.61/21 8} Rbd8 {0.21/15 8} 39.
Rg3 {(Qc1-b1) 0.69/20 1} Qf6 {0.49/16 6} 40. Ng2 {0.62/20 2} Ne6 {0.49/16 10}
41. Qe3 {0.62/20 17} Rb8 {(Ne6-c5) 0.54/14 4} 42. Be2 {0.72/20 8} Nd4 {0.48/15
8} 43. Qh6+ {(Ng2-h4) 0.77/19 3} Ke7 {0.34/13 2} 44. Ne1 {(Ng2-h4) 0.77/19 2}
Rbd8 {0.39/15 16} 45. Nf3 {0.79/19 2} Ne6 {0.56/15 3} 46. Qe3 {0.79/19 1} c5 {
0.46/14 5} 47. Nd2 {(Be2-d3) 1.02/19 6} Nf4 {0.57/16 10} 48. Nb1 {0.93/19 2}
Qd6 {(Qf6-c6) 0.57/16 4} 49. Nc3 {1.05/19 1} Qd4 {0.57/16 3} 50. Qxd4 {1.09/18
1} Rxd4 {0.87/16 2} 51. Bd3 {(Nc3-d5+) 1.09/18 1} Rh8 {(Re8-f8) 0.67/16 5} 52.
Nd5+ {(Rg3-f3) 1.17/19 4} Nxd5 {0.72/15 1} 53. exd5 {1.27/21 4} Kd6 {(h5-h4) 0.
82/17 3} 54. f3 {(Rg3-e3) 1.38/20 1} Rf4 {0.86/17 3} 55. Kg2 {(Kf1-f2) 1.38/21
1} Rd4 {(Rh8-d8) 0.89/18 3} 56. Rb1 {1.38/21 1} Rb8 {0.89/19 1} 57. Rg5 {
(Kg2-f2) 1.38/22 1} Rf4 {(Rd4-h4) 0.99/20 10} 58. Rh1 {(Kg2-f2) 1.38/23 4} Rh8
{(Rb8-e8) 0.99/18 3} 59. Re1 {1.38/23 4} Re8 {0.99/18 0} 60. Ra1 {(Bd3-e4) 1.
38/23 1} Rg8 {(Rf4-d4) 0.99/18 3} 61. Ra2 {(Kg2-g3) 1.38/22 1} Rh4 {(Rg8-b8) 0.
99/20 4} 62. Rb2 {1.38/22 1} Rb8 {0.99/20 0} 63. Rb3 {1.38/23 1} Rd4 {0.99/20 1
} 64. Rb5 {(Kg2-g3) 1.38/23 1} Rh4 {(Rd4-f4) 0.99/21 3} 65. Be4 {(Rb5-b2) 1.38/
21 1} Rb7 {(Rh4-f4) 0.99/18 3} 66. Rb1 {(Rb5-b3) 1.38/24 4} Rb8 {1.02/20 2} 67.
Ra1 {(Kg2-g3) 1.38/23 1} Rg8 {(Rh4-f4) 1.02/19 3} 68. Ra2 {(Kg2-f2) 1.38/23 3}
Rb8 {1.02/21 3} 69. Ra3 {(Ra2-b2) 1.38/23 1} Rf4 {(Rb8-g8) 1.02/20 2} 70. Re3 {
1.38/19 1} Rh4 {(Rb8-e8) 1.02/19 1} 71. Re1 {(Re3-b3) 1.38/20 1} Rg8 {(Rh4-f4)
1.02/20 2} 72. Rg3 {(Re1-a1) 1.38/22 5} Rf4 {1.02/16 2} 73. Rb1 {1.38/23 1} Rb8
{1.02/17 0} 74. Kf2 {1.38/22 1} Rh4 {(Rf4-f6) 1.02/17 1} 75. Rd1 {1.38/21 1}
Rh2+ {1.02/18 1} 76. Kg1 {1.38/23 1} Rh4 {1.02/19 0} 77. Rg5 {1.38/23 1} Rf4 {
(Rh4-h3) 1.02/19 1} 78. Re1 {(Kg1-f2) 1.38/21 1} Rg8 {(Rb8-d8) 1.02/18 1} 79.
Rb1 {1.38/24 3} Rb8 {1.02/19 0} 80. Kf2 {1.38/24 2} Ke7 {(Rb8-b7) 1.02/18 1}
81. d6+ {(Rg5-g1) 1.54/21 4} Kf6 {(Ke7xd6) 1.36/15 2} 82. Rgg1 {1.54/22 2} Ke6
{1.35/16 1} 83. Bd5+ {1.62/22 4} Kxd6 {1.39/16 1} 84. Rg5 {1.85/23 12} Rf5 {
(Rf4-f6) 1.39/16 3} 85. Rxf5 {1.96/21 1} gxf5 {1.39/16 0} 86. Bxf7 {1.96/21 1}
h4 {1.54/16 0} 87. Kg2 {2.08/21 1} Ke7 {1.54/16 2} 88. Bd5 {2.16/22 0} Kf6 {1.
84/18 1} 89. Kh3 {(f3-f4) 2.23/22 0} Kg5 {2.05/16 0} 90. Rg1+ {2.23/22 0} Kf4 {
(Kg5-f6) 2.05/17 0} 91. Rg6 {2.38/18 0} Re8 {(Kf4-e3) 2.05/18 1} 92. Rf6 {
(Rg6-h6) 2.65/23 3} Kg5 {2.05/15 1} 93. Rxb6 {2.78/22 1} e4 {(Kg5-f4) 2.45/14 1
} 94. fxe4 {3.76/18 1} fxe4 {3.16/16 1} 95. Re6 {3.94/19 0} Rxe6 {(Re8-b8) 3.
16/18 0} 96. Bxe6 {4.47/18 0} Kf6 {4.45/18 2} 97. Bd7 {(Be6-g4) 4.56/20 0} e3 {
(Kf6-e5) 1.66/14 0} 98. Bg4 {6.53/24 1} Ke5 {(Kf6-g5) 1.66/15 1} 99. Kxh4 {
(c2-c3) 8.22/24 2} Kd4 {(Ke5-f4) 1.97/13 0} 100. Kg3 {10.71/24 2} Kxc4 {
(Kd4-e4) 2.72/14 1} 101. Kf3 {198.30/25 1} Kd4 {(Kc4-b4) 4.45/13 0} 102. Bh5 {
(Bg4-f5) 198.89/24 0} Kc3 {(e3-e2) 4.46/15 1} 103. Kxe3 {201.45/27 1} Kxc2 {
(c5-c4) 4.46/15 0} 104. Ke4 {(Bh5-e2) 219.90/0 0} Kb3 {4.46/15 1} 105. Bd1+ {
219.91/0 0} Kc4 {#21/14 0} 106. Ke5 {219.92/0 0} Kc3 {#19/13 0} 107. Kd5 {219.
93/0 0} Kb4 {#19/15 0} 108. Bc2 {219.94/0 0} c4 {#18/16 1} 109. Kd4 {(Kd5-c6)
219.95/0 0} c3 {#17/18 1} 110. Bd1 {219.96/0 0} Ka3 {Time #17/17 0} 0-1

Here is the problem game.Ivanhoe has fought long and hard and has forced mate in 17 moves.Instead to finish the game in the last few seconds left it stops playing and loses on time.
Maybe the compilers and programmers can conclude something from this game.
Also there were additional few games played on AMD computer where Ivanhoe was causing problems but Rybka was losing on time so something is still wrong,maybe just with this compile.

Rendruk
29.07.2010, 23:57
Yes, it's a well known fact, but this works for people, not for software.

You might want to read up on "Cognitive Dissonance"

Wikipedia's has a good reference article that can link to peer-reviewed scholarly writings if you wanted to learn more than the summarization of the concept. Also check up on "Synthesized Happiness". It's all related to each other and the original concept you said you knew of. If you're into that sort of thing it's interesting.

I'll check out that driver update extension, although I've used lots of driver update utilities before and it rarely finds one I need. Mostly new keyboard/mouse driver updates.

Vlad0
30.07.2010, 00:10
Here is the problem game.Ivanhoe has fought long and hard and has forced mate in 17 moves.Instead to finish the game in the last few seconds left it stops playing and loses on time.
Maybe the compilers and programmers can conclude something from this game.


This is really strange. On my note (slow Core 2 Duo) it suffices for 52bU
even a fraction of a second to finish <= 5 pirces endspiel when total
bases are registered. And they are registered and are working
in this case as the evaluation 229 shows. And here there was even several seconds.
The description of this bug resembles wery much the "sleeping bug" present in the original code
of v.53. But in the original code of v.52 this bug was already fixed. Yet it looks that some traces
of it still are hidden somewhere. One is to find the place -- where.

jc.m
30.07.2010, 00:43
This is really strange. On my note (slow Core 2 Duo)
....
The description of this bug resembles wery much the "sleeping bug" present in the original code
of v.53. But in the original code of v.52 this bug was already fixed. Yet it looks that some traces
of it still are hidden somewhere. One is to find the place -- where.

Yes, but in this case, it's an AMD processor.
I'm quite glad that the bug I mentioned was reproduced.
In my case, I get it sometimes when I run "benchmark" with multiprocessor.
Here is how I get it (sometimes):
Type:
setoption name TitanicMode value true
setoption name TitanicCPUsSizings value 2
benchmark

and type benchmark several times to get it.
At one moment, it freezes during a few seconds.
Probably a race condition (these are the worst bugs to find).
Also, from my memory, it happens with IC++, not VC++ builds.

jc.m
30.07.2010, 01:48
I found a bug in Ivanhoe, while analyzing a game between Portisch and Fischer (and I mistyped the game).
Save the following line into a file:
position startpos moves d2d4 g8f6 c2c4 e7e6 b1c3 f8b4 e2e3 b7b6 g1e2 c8a6 e2g3 b4c3 b2c3 d7d5 d1f3 e8g8 e3e4 d5e4 g3e4 f6e4 f3e4 d8d7 c1a3 f8e8 f1d3 f7f5 e4a8 b8c6 a8e8 d7e8 e1g1 c6a5 a1e1 a6c4 d3c4 a5c4 a3c1 c7c5 d4c5 c5b6

then run:
Ivanhoe <thisfile

it crashes into utility.c, ERROR_END on vfprintf.
In fact, we must move the line va_end just after the vfprintf, otherwise it passes a NULLPTR and crashes badly.
Please, could you report this bug to the Decembrists ? (I'll never post on their forum again).
Thanks !

slankamen
30.07.2010, 01:55
Intel(R) Core(TM)2 Quad Q9550 2.83GHz4x @3500 MHz 4,096 MB Memory
Microsoft Windows XP 64 Bit Professional Service Pack 2 (Build 3790)
Fritz Benchmark:
Speed: 20.90
KNS: 10032
GUI: CB Rybka 3
Book: Perfect 2009 10 moves
Hash: 256
RB and TB: ON
Ponder: OFF
Blitz:10' 0


1 Ivanhoe 52bU-x64 3185 +8/-4/=18 56.67% 17.0/30
2 Deep Rybka 4 x64 3190 +4/-8/=18 43.33% 13.0/30

This version by Vlad0 looks very strong and this is second match where it was winning against Rybka 4.Now will test against Houdini 1.03a.x64.
:sm36:

Rendruk
30.07.2010, 02:05
I really need some help.

Could someone doublecheck my findings, which are:

EnableLargePages actually hurts ELO.

Twice now I've run STS tests and gotten 100 points taken off whenever that option was enabled.

Want someone to see if that happens to them too. Thank you.

BTO7
30.07.2010, 02:23
Woah, you are right !
I didn't think it was possible to be this lame, but the author binary edited the files. His name is "Luis Olmos" (note the 3 spaces between his first and last names :D).
Also, the compilation path is: R:\Schaak\Tactico\VS\4MP\Release\Tactico _64.pdb
And Schaak means chess in dutch.

This is the same author of the DeepGO engine. I still have the file. I obtained it from the Angels site and they know this person. They seem to be friends. DeepGO is in another thread here with the same modifications done...except it was a rip off of Deep Rybka. So apparently this is what this guy does regularly.

Regards
BT