Stefan Pohl Computer Chess

Latest Website-News (2024/11/20): Testrun of Ceres 1.01 C1-512-25 finished (Ceres runs without additional TensorRT files (which give an additionally speedup), because Lc0 does not use these files and so, the results of Lc0 and Ceres are better comparable, when using the "default"-Ceres). Mention, in TCEC with it's fast hardware and long timecontrols, a speedup makes no big Celo-difference, but in my testings, a speedup of Ceres, which Lc0 does not get, will lead to heavy distortions): 

See the result and download the played games on NN vs Stockfish 15.1 testing

 

Dont forget to take a look at my EAS-Ratinglist (The world's first engine-ratinglist not measuring strength of engines, but engines's style of play)

 

Next NN-testrun: Ceres 1.01 C1-640-25 net (Ceres nets downloads here)

 

 

Stay tuned

UHO-Top15 Engines Ratinglist (+ regular testing of Stockfish Dev-versions)

 

 

Playing conditions:

 

The 15 strongest engines and the latest StockfishDev version are playing 1000 games vs. each opponent: 15000 games per engine are played. A RoundRobin Tournament with 120000 games overall !

 

Hardware: AMD Ryzen 9 7945HX 16-core (32 threads) notebook with 64GB RAM (since 2024/10/08), 2 identical machines.

Speed: (singlethread, TurboBoost-mode switched off, chess starting position) Stockfish 17: 600 kn/s (when 26 games are running simultaneously)

Hash: 512MB per engine

GUI: Cutechess-cli (GUI ends game, when a 5-piece endgame is on the board, all other games are played until mate or draw by chess-rules (3fold, 50-moves, stalemate))

Tablebases: None for engines, 5 Syzygy for cutechess-cli

Openings: My UHO_2022_6mvs_+120_+129 openings are used (first 500 lines) (UHO 2022 openings are part of my Anti Draw Openings download-package). 

Ponder, Large Memory Pages & learning: Off

Thinking time: 3min+1sec per game/engine. One testrun (15000 games) takes around 80-85 hours (=3.5 days)

The version-numbers of the Stockfish engines are the date of the latest patch, which was included in the Stockfish sourcecode, not the release-date of the engine-file, written backwards (year,month,day))(example: 200807 = August, 7, 2020). The used SF compile is the AVX512-compile, which is the fastest on my AMD Ryzen CPU. SF binaries are taken from https://github.com/official-stockfish/Stockfish/releases (except the official SF-release versions, which are taken form the official Stockfish website).

Read the explanation, why I chose my testing-conditions, like I did here

 

Latest update: 2024/11/18: Stockfish 241113  (-2 Celo to Stockfish 241031) and Lizard 11.1 (Lizard re-entered and kicked Stormphrax 6.0.0 out)

 

(Ordo-calculation fixed to Stockfish 17 = 3843 Celo. Mention, there are no human or "realistic" Elo numbers in enginechess, since the engines reached superhuman strength, so I decided to use "Celo" (=Computer Elo) instead)

 

See the tournament cross-table as a picture (done by Fritz18 GUI) here

See the individual statistics of engine-results here

See the Engines Aggressiveness Score Ratinglist here

See the Gamepair-statistics of the UHO-Top15 Ratinglist here

See the full UHO Ratinglist (with EAS-Ratinglist): here

Download the current gamebase, the full-list gamebase and the archive here

See the old SPCC-Ratinglist (full list) and EAS-List from 2020 until 2023/08/28: here

 

(best Stockfish Celo: Stockfish 241004: 3850 Celo, latest full release (Stockfish 17) had 3843 Celo)

(highest EAS-Score since Stockfish 16 release: Stockfish 240519: 203882 EAS-points)

(Read the explanation of the Lc0-entry in the Ratinglist here).

     Program                    Celo    +    - Games    Score   Av.Op. Draws

   1 Stockfish 241113 a512    : 3847    4    4 15000    70.0%   3695   48.9%
   2 Stockfish 17 240906      : 3843    4    4 15000    69.5%   3695   48.8%
   3 Torch 3 popavx2          : 3801    4    4 15000    64.0%   3698   49.2%
 [   Lc0 0.32dev BT4-6077500  : 3764   15   15  1000 (NN vs SF15.1 testing)]
   4 Obsidian 14.0 a512       : 3746    4    4 15000    56.2%   3701   50.3%
   5 Berserk 13 avx2          : 3722    4    4 15000    52.8%   3703   51.2%
   6 KomodoDragon 3.3 avx2    : 3719    4    4 15000    52.3%   3703   50.9%
   7 Caissa 1.21 a512         : 3701    4    4 15000    49.7%   3704   50.5%
   8 Alexandria 7.1 a512      : 3696    4    4 15000    48.9%   3705   50.9%
   9 PlentyChess 2.1 avx2     : 3671    4    4 15000    45.3%   3706   51.1%
  10 Ethereal 14.38 avx2      : 3668    4    4 15000    44.8%   3707   50.4%
  11 RubiChess 240817 avx2    : 3653    4    4 15000    42.7%   3708   50.3%
  12 Starzix 6.0 a512         : 3651    4    4 15000    42.4%   3708   50.6%
  13 Viridithas 15.0 a512     : 3644    4    4 15000    41.4%   3708   50.7%
  14 Integral 5.0 a512        : 3643    4    4 15000    41.3%   3708   49.0%
  15 Clover 8.0 avx2          : 3638    4    4 15000    40.5%   3709   50.9%
  16 Lizard 11.1 a512         : 3622    4    4 15000    38.2%   3710   49.0%

Games        : 120000 (finished)

White Wins   : 58361 (48.6 %)
Black Wins   : 1447   (1.2 %)
Draws        : 60192 (50.2 %)

Below the UHO-Top15 Ratinglist, recalculated with Gamepairs (using my Gamepair Rescoring Tool V1.5), Ordo-calculation fixed to Stockfish 17 = 3843 Celo, realizing Vondele's (Stockfish Maintainer) idea: "Thinking uniquely in game pairs makes sense with the biased openings used these days. While pentanomial makes  sense it is a bit complicated so we could simplify and score game pairs only (not games) as W-L-D (a traditional score of 2-0, or 1.5-0.5 is just a W)."

See the head-to-head gamepair-statistics of the UHO-Top15 Ratinglist here

   # PLAYER                   :    Celo  Error   Pairs     W     D     L   (%)  CFS(%)
   1 Stockfish 241113 a512    :    3856     12    7500  5973  1371   156  88.8      98
   2 Stockfish 17 240906      :    3843   ----    7500  5888  1426   186  88.0     100
   3 Torch 3 popavx2          :    3693     12    7500  4685  2150   665  76.8     100
   4 Obsidian 14.0 a512       :    3547     12    7500  3233  2795  1472  61.7     100
   5 Berserk 13 avx2          :    3493     12    7500  2699  2909  1892  55.4      91
   6 KomodoDragon 3.3 avx2    :    3488     11    7500  2582  3043  1875  54.7     100
   7 Caissa 1.21 a512         :    3444     11    7500  2135  3106  2259  49.2     100
   8 Alexandria 7.1 a512      :    3432     12    7500  2005  3140  2355  47.7     100
   9 PlentyChess 2.1 avx2     :    3377     12    7500  1492  3119  2889  40.7      94
  10 Ethereal 14.38 avx2      :    3370     12    7500  1441  3097  2962  39.9     100
  11 RubiChess 240817 avx2    :    3346     12    7500  1289  2943  3268  36.8      99
  12 Starzix 6.0 a512         :    3336     11    7500  1177  2986  3337  35.6     100
  13 Viridithas 15.0 a512     :    3321     11    7500  1042  2981  3477  33.8      98
  14 Integral 5.0 a512        :    3313     11    7500  1000  2912  3588  32.7      99
  15 Clover 8.0 avx2          :    3302     12    7500   937  2847  3716  31.5     100
  16 Lizard 11.1 a512         :    3261     12    7500   670  2679  4151  26.8     ---

 

------------------------------------------------------------------- 
--- Number of all Gamepairs          : 60000 
--- Number of drawn Gamepairs overall: 21752 (= 36.25%) 
--- Number of 1:1 drawn Gamepairs    : 10277 (= 17.13%) 
--- Number of 2-draws drawn Gamepairs: 11475 (= 19.12%) 
------------------------------------------------------------------- 

Aborted testruns, because the tested engine was too weak to enter the UHO-Top15 Ratinglist:

Minic 3.41, Booot 7.3, Black Marlin 9.0, Equisetum 1.0

Below you find a diagram of the progress of Stockfish in my tests since February 2023. 

And below that diagram, the older diagrams.

 

You can save the diagrams on your PC with mouseclick (right button) and then choose "save image"...