Rybka 3 - Che 2.1 match, 8CPU

[Gusev]

I am broadcasting a 180-game match between Rybka 3 and Che 2.1 (a clone of Igorrit 0.086v9) at http://foscec.org/RybkaChe/index.htm. The games will be available at http://foscec.org/RybkaChe/games.pgn. The time control is 30'+30". The number of threads is 8. (This Rybka 3 is the x64 Deep version, so it does support up to 8 threads. I acquired it new at Amazon for $49.99.) Hash size: 2048M. Processor: Intel Core i7-5960X.

[Gusev]

I am broadcasting a 180-game match between Rybka 3 and Che 2.1 (a clone of Igorrit 0.086v9) at http://foscec.org/RybkaChe/index.htm. The games will be available at http://foscec.org/RybkaChe/games.pgn. The time control is 30'+30". The number of threads is 8. (This Rybka 3 is the x64 Deep version, so it does support up to 8 threads. I acquired it new at Amazon for $49.99.) Hash size: 2048M. Processor: Intel Core i7-5960X.

Rybka is ahead, +3-0=8, after the first 11 games. Che caused an exception during Game 3.

[Gusev]

Two more exceptions were caused by Che, so it looks like I should treat these cases separately in order to avoid reaching a wrong conclusion as to the relative strength of the two engines. After 20 games, 17 of them were completed, and the score is even, +2-2=13.

[kranium]

Hi Dimitri-
not sure what compiler you're using, but you might want to try less aggressive (or no) compiler optimizations...

for ex: start with a generic MS compile, and run it overnight to verify stability
if the app doesn't crash, then proceed to add an optimization, test again etc.

If you don't have time or motivation for this, you can probably find an old 2009/2010 compile (of Igorrit 0.086v9) that is stable.

[Gusev]

Hi Dimitri-
not sure what compiler you're using, but you might want to try less aggressive (or no) compiler optimizations...

for ex: start with a generic MS compile, and run it overnight to verify stability
if the app doesn't crash, then proceed to add an optimization, test again etc.

If you don't have time or motivation for this, you can probably find an old 2009/2010 compile (of Igorrit 0.086v9) that is stable.

Hi Norm,

I have Sentinel's old compile, but it did crash on one occasion under Arena on my other computer. These crashes of the new compile are happening under the auspices of Deep Fritz 14. I also have a generic MSVC compile of Che, so I will test to find out how stable it is. However, its test under Arena showed that it was ~-14 Elo compared to the optimized compile. Some other factor may be in play here, too: During FOSCEC Season 2, OpenCritter 1.1.37 caused an exception twice, and the (consequently unreleased) Ivanhoe 1945b did so once. OpenCritter was compiled from its ObjectPascal code using Embarcadero Studio XE7 without any changes.

[Gusev]

Hi Dimitri-
not sure what compiler you're using, but you might want to try less aggressive (or no) compiler optimizations...

for ex: start with a generic MS compile, and run it overnight to verify stability
if the app doesn't crash, then proceed to add an optimization, test again etc.

If you don't have time or motivation for this, you can probably find an old 2009/2010 compile (of Igorrit 0.086v9) that is stable.

Hi Norm,

I have Sentinel's old compile, but it did crash on one occasion under Arena on my other computer. These crashes of the new compile are happening under the auspices of Deep Fritz 14. I also have a generic MSVC compile of Che, so I will test to find out how stable it is. However, its test under Arena showed that it was ~-14 Elo compared to the optimized compile. Some other factor may be in play here, too: During FOSCEC Season 2, OpenCritter 1.1.37 caused an exception twice, and the (consequently unreleased) Ivanhoe 1945b did so once. OpenCritter was compiled from its ObjectPascal code using Embarcadero Studio XE7 without any changes.

The match did not end particularly well. Che won, 74:64, in the 138 games that were completed, but crashed in the other 42 games. The compile used must have been too optimized. Also, it was not optimized for the AVX2 architecture of the processor used, but for the more common SSE4.2.