Leela Chess Zero 42703 vs Stockfish 010719 CCRL(f) = 28h/40

[mwyoung]

Leela Chess Zero 42703 vs Stockfish 010719 CCRL(f) = 28h/40

Time control: Equivalent to 40 moves in 28 hours on Athlon 64 X2 4600+ (2.4 GHz) x 16 cores

Hardware 2950x, RTX 2080 ti
Stockfish 32 Thread, 16 Gb HT
Leela Chess Zero 4 threads, 9900000 NN CS.
6 man Tablebase
8 move opening book
TC= 40m/40
Ponder = On

Live Stream: https://www.youtube.com/watch?v=iadHV2Jpvrs

[mwyoung]

Update Live Stream: https://www.youtube.com/watch?v=ejffOO2rOnE

[mwyoung]

Leela Chess Zero 42703 vs Stockfish 010719 CCRL(f) = 28h/40

Time control: Equivalent to 40 moves in 28 hours on Athlon 64 X2 4600+ (2.4 GHz) x 16 cores

Hardware 2950x, RTX 2080 ti
Stockfish 32 Thread, 16 Gb HT
Leela Chess Zero 4 threads, 9900000 NN CS.
6 man Tablebase
8 move opening book
TC= 40m/40
Ponder = On

Live Stream: https://www.youtube.com/watch?v=iadHV2Jpvrs

DESKTOP-CORSAIR, 40m/40+40m/40+40m/40 0


1 Stockfish 010719 64 POPCNT +17 +3/=15/-2 52.50% 10.5/20
2 Lc0 v0.21.2 -17 +2/=15/-3 47.50% 9.5/20

[jorose]

Are you saying you are getting 42x single threaded speedup over single threaded performance of the Athlon when running a 32bit program like the Crafty Benchmark? Because a quick google search makes me think it should be more like 3x.

https://www.cpubenchmark.net/cpu.php?cp ... 0%2B&id=81

https://www.cpubenchmark.net/cpu.php?cp ... 0X&id=3316

[RubiChess]

The guy mentioned cinebench(!) for comparing system speed in one of his threads. So we probably should take it as it is and stop refreshing this nasty discussion.