Stefan Pohl Computer Chess

private website for chessengine-tests


Long thinking-time testing of LC Zero CUDA

 

Playing conditions:

 

Hardware: i7-6700HQ (2.6 GHz) Notebook (Quadcore), Windows 10 64bit, 12GB RAM

Fritzmark: 4 cores / 7 threads: 18.4 / 8825 (CPU-speed reduced to 99% (to switch off the Intel Turbo Boost))

GPU (used by LC Zero): Nvidia Cuda Gforce GTX 950M (4GB memory, grafics: 914 MHz, memory: 1.25 GHz). LC Zero CUDA calculates around 1020 rollouts/s in the starting position (measured with "go infinte"-command until depth 28) with Netsize 20x256

Leela-Ratio: 0.67 for my tournament (opponent engines running singlethread)(what is it? look here) (The Ratio-value of AlphaZero in the match vs. Stockfish 8 was 1.0, and the Ratio-value in TCEC is around 0.35) so a value of 0.67 is very close (and between) to both tournaments.  Better GPUs (like GTX 1080) have much higer Ratio-Values (6 or higher), which means a huge advantage for Leela, when playing vs. other engines and that leads to distorted results (Leela seems to be stronger, than it is).)

Hash: 2 GB per engine

GUI: Fritz 16 GUI (no draw, no resign)

Tablebases: 3-6 Syzygy for GUI (GUI ends all games, when only 6 pieces remain on the chessboard)

Openings: 4 moves opening book. Download it here

Large Memory Pages: Off

Ponder: Off

Thinking time: 12'+5'' per game/engine (average game-duration: 30'-40', so around 40-45 games per day are played). Average thinking time in the middlegame is around 20'' per move.

 

LC Zero Github (information, download Networks and LC Zero Engine): here

Read more about, how LC Zero works, in the LC0-Blog: here

See a step-by-step manual for using LC Zero CUDA in FritzGUI and ShredderGUI here

 

Current participants: None. Tournament suspended, because the neural-net learning of LC Zero was restarted and until now, the new 20xxx nets are much weaker, than the latest 10xxx nets. My testruns will continue, when the new nets are getting as strong as the older 10xxx nets. I will do some fast testruns of LC Zero vs. Stockfish from time to time, to measure the progress of the 20xxx nets...

 

10 opponents:

Stockfish 9 (Elo 3380), Houdini 6 (Elo 3349), Komodo 11.3 (Elo 3329), Fire 7.1 (Elo 3233), Booot 6.3.1 (Elo 3166), Shredder 13 (Elo 3156), Fizbo 2 (Elo 3134), Ethereal 10.55 (Elo 3132), Andscacs 0.94 (Elo 3122), Gull 3 (Elo 3078)

 

now with LC Zero CUDA v0.17.0-rc2 (2 worker threads)

 

Average opponent CEGT Elo (40moves/20', singlethread): 3208

All opponent-engines running with default settings and in singlethread-mode, like in CEGT-testruns.

Leelas time-management is still too slow in the early middlegame. So, I reduced Scale thinking time from 2.4 to 2.0.

 

Latest update: 2018/09/05 - download all games (4600 games): here

 

 

LC0 Cuda v0.17.0 N11141 (size: 20x256)(date:180827): 320 (+ 97,=170,- 53)
56.9 % (opponent-Elo: 3208) Elo-performance: 3256

 

LC0 Cuda v0.16.0 N10860 (size: 20x256)(date:180818): 320 (+ 70,=171,- 79)
48.6 % (opponent-Elo: 3208) Elo-performance: 3198
 

LC0 Cuda v0.16.0 N10555 (size: 20x256)(date:180809): 320 (+ 91,=175,- 54)
55.8 % (opponent-Elo: 3208) Elo-performance: 3248

 

LC0 Cuda v0.16.0 N 529 (size: 15x192)(date: 180731): 300 (+ 60,=133,-107)
42.2 % (opponent-Elo: 3135) Elo-performance: 3080

 

LC0 Cuda v0.16.0 N 512 (size: 15x192)(date: 180722): 300 (+122,=129,- 49)
62.2 % (opponent-Elo: 3000) Elo-performance: 3087

 

LC0 Cuda(180711) N 492 (size: 15x192)(date: 180714): 300 (+100,=137,- 63)
56.2 % (opponent-Elo: 3000) Elo-performance: 3043

 

LC0 Cuda(180619) N 473 (size: 15x192)(date: 180707): 300 (+101,=128,- 71)
55.0 % (opponent-Elo: 3000) Elo-performance: 3035

 

LC0 Cuda(180619) N 452 (size: 15x192)(date: 180629): 320 (+101,=136,- 83)
52.8 % (opponent-Elo: 3000) Elo-performance: 3019

 

LC0 Cuda(180619) N 432 (size: 15x192)(date: 180622): 300 (+ 85,=138,- 77)
51.3 % (opponent-Elo: 3000) Elo-performance: 3009

 

LC0 Cuda(180602) N 374 (size: 15x192)(date: 180604): 300 (+ 86,=114,-100)
47.7 % (opponent-Elo: 3000) Elo-performance: 2984

 

LC0 Cuda(180530) N 359 (size: 15x192)(date: 180530): 300 (+ 79,=113,-108)
45.2 % (opponent-Elo: 3000) Elo-performance: 2967

 

LC0 Cuda(180521) N 322 (size: 15x192)(date: 180520): 300 (+ 68,=116,-116)
42.0 % (opponent-Elo: 3000) Elo-performance: 2944

 


lczero v0.8 (GPU) N 251 (size: 15x192)(date: 180505): 320 (+164,= 73,- 83)
62.7 % (opponent-Elo: 2524) Elo-performance: 2614


lczero v0.7 (GPU) N 214 (size: 10x128)(date: 180429): 300 (+157,= 85,- 58)
66.5 % (opponent-Elo: 2524) Elo-performance: 2643


lczero v0.7 (GPU) N 162 (size: 10x128)(date: 180421): 300 (+113,= 66,-121)
48.7 % (opponent-Elo: 2524) Elo-performance: 2515