SPCC: Testrun of Torch 3.1 finished

[pohl4711]

My UHO-Top15 Ratinglist is the world's first engine-ratinglist, using UHO-openings, and the world's first ratinglist offering additionally Gamepair-statistics.

Testrun of Torch 3.1 (Torch 3 with a memory-speedup-patch) finished: +8 Celo to Torch 3.
Additionally I did re-test of Torch 3 with +30% more thinking-time for Torch 3, for a comparison of the Celo-gains of the memory-speedup on one hand and the +30% increased thinking-time on the other hand. Thanks to Andrew Grant for providing me Torch 3.1 and allowing me to make these interesting experiments! You find the result on the Experiments site: Here Torch 3 gained +20 Celo. So, if +30% more speed/time leads to a Celo gain of +20 and Torch 3.1 gained +8 Celo in it's testrun, the Torch memory-speedup patch should lead to speedup of (around) +12% in my testing-environment.

https://www.sp-cc.de

Also take a look at the EAS-Ratinglist, the world's first engine-ratinglist not measuring strength of engines but engines's style of play:
https://www.sp-cc.de/eas-ratinglist.htm

(Perhaps you have to clear your browsercache with <strg>+<shift>+<delete> to reload the graphics/diagrams on my website)

[pohl4711]

As usual, the rating in the full ratinglist differs a lot, because Torch 3 played way more games (45000) comapred to Torch 3.1 (14000): +19 Celo here in the full ratinglist - only +8 Celo in the UHO-Top15 (RoundRobin) ratinglist...

Code: Select all

     Program                     Celo    +    - Games    Score   Av.Op. Draws

   1 Stockfish 17 240906       : 3843    3    3 37000    71.2%   3679   48.6%
   2 Torch 3.1 a512            : 3825    4    4 14000    65.7%   3709   48.6% (NEW)
   3 Stockfish 16.1 240224     : 3810    2    2 55000    72.5%   3632   47.1%
   4 Torch 3 popavx2           : 3806    3    3 45000    68.6%   3664   47.6% (OLD)
   5 Torch 2 popavx2           : 3787    3    3 38000    71.1%   3624   47.3%
   6 Stockfish 16 230630       : 3782    3    3 39000    74.4%   3586   45.1%

But what is really interesting is, that the EAS-score of both versions (which are playing identical except Torch 3.1 is running +12% faster) is nearly identical, despite the huge difference of played games, and the non-identical opponents:

Code: Select all

                                 bad  avg.win 
Rank  EAS-Score  sacs   shorts  draws  moves  Engine/player 
-------------------------------------------------------------------
  25    119408  13.52%  19.92%  14.76%   72   Torch 3.1 a512  
  26    118296  12.66%  19.63%  14.11%   72   Torch 3 popavx2

Great proof, how good my EAS-Tool works, if the number of played games per engine is 10000 or higher. Just 1100 points difference, means less than 1% difference in EAS-scoring. Nice!

[AndrewGrant]

Thanks for testing, Stefan.

I expect to see rapid adoption by the community, which was the ultimate goal. I've been very open about the process, and even helped share needed mmap code, and knowledge about the pre-processing to get it done. The end goal is to force Stockfish to adopt the standard.

We'll see. If/when I return to Ethereal, I'll be doing this and publishing the tools for it. Hopefully by then I still get some attribution.