TalkChess.com

Stockfish 141020 (without net) vs Stockfish 310720 Program Elo + - Games Score Av.Op. Draws 1 Stockfish 310720 x64 bmi2 : 2403 12 12 1200 51.0 % 2397 60.7 % 2 Stockfish 141020 x64 bmi2 : 2397 12 12 1200 49.0 % 2403 60.7 % Individual statistics: 1 Stockfish 310720 x64 bmi2 : 2403 1200 (+248,=728,-22...

BrainLearn vs Stockfish Match Program Elo + - Games Score Av.Op. Draws 1 Stockfish 141020 x64 bmi2 : 2406 9 9 1200 51.6 % 2394 79.3 % 2 BrainLearn 12.1 x64 bmi2 : 2394 9 9 1200 48.4 % 2406 79.3 % Individual statistics: 1 Stockfish 141020 x64 bmi2 : 2406 1200 (+143,=952,-105), 51.6 % BrainLearn 12.1...

Nets trained by James Horsfall Thomas(Jhortos) is really amazing.
Number of trained games of 320 x 24 net is ~130.000.000. This number is ~3x of number of trained games of AlphaZero.

ShashChess vs Stockfish Match Program Elo + - Games Score Av.Op. Draws 1 Stockfish 141020 x64 bmi2 : 2411 11 11 1000 53.1 % 2389 74.4 % 2 ShashChess 15.1 x64 bmi2 : 2389 11 11 1000 46.9 % 2411 74.4 % Individual statistics: 1 Stockfish 141020 x64 bmi2 : 2411 1000 (+159,=744,- 97), 53.1 % ShashChess ...

Leo wrote: ↑

Tue Oct 06, 2020 4:57 pm

I never heard of the engine. Maybe it jumped in power because of NN only. I get the feeling it wont progress much now. I hope it will.

Nemorino has been participating in the TCEC tournaments which has been continuously since TCEC Season10.

Amazing performance of Nemorino 6.
Has Nemorino lost its originality because it uses its own nnue network? I don't think so.

Cfish vs Stockfish Match Program Elo + - Games Score Av.Op. Draws 1 Cfish 031020 x64 bmi2 : 2403 7 7 1540 50.8 % 2397 83.2 % 2 Stockfish 280920 x64 bmi2 : 2397 7 7 1540 49.2 % 2403 83.2 % Individual statistics: 1 Cfish 031020 x64 bmi2 : 2403 1540 (+142,=1282,-116), 50.8 % Stockfish 280920 x64 bmi2 ...

Amazing difference with NNUE net. +300 elo is an incredible number. These NNUE net's are breaking the chess lists.
It will be very difficult in the near future to see chess engines owned by programmers who refuse to utilize these networks at the top the rating lists.

Laskos wrote: ↑

Mon Sep 28, 2020 4:18 pm

What is "baseline" and "optimized"?

https://github.com/LeelaChessZero/lc0/pull/1428

Lc0 benchmarks with SV-3010 network (384x30) Default settings (minibatch-size=256) --------------------------------------------- GPU baseline optimized perf gain (%) --------------------------------------------- Titan RTX.. 17443 - 20084 15.1 RTX 3090.. 26820 - 29767 11.0 A100........ 41785 - 48815...