Jhoravi wrote: ↑Thu Sep 03, 2020 3:24 pm
Is Hyperthreading preferable in NNUE?
Good question, most likely yes, As always it needs to be tested on your system.
The more physical cores on your system. Usually the more you need to watch out for over threading.
8 cores could work well with 16 threads. But if you have 64 cores. It may not be the best going to 128 threads...
"The worst thing that can happen to a forum is a running wild attacking moderator(HGM) who is not corrected by the community." - Ed Schröder
But my words like silent raindrops fell. And echoed in the wells of silence.
Stockfish 12 still cannot solve the win study [d]2R2K2/p3p3/3p4/p2P4/P7/4p3/6pp/6nk w - - 0 1 within a couple of minutes (4 cores, Mac, 8GB hash via brew). Even after the introductory moves 1. Rc1 e2, Stockfish 12 did not seem to see the winning move 2 Rb1! after a while.
I wish these engines were tuned better for the analysis of studies.
lkaufman wrote: ↑Thu Sep 03, 2020 1:15 am
My final result (SF 12 vs 11, 1 thread, 2' + 1" adapted by CCRL rules for hardware) was 180 wins, 19 losses, 301 draws, which is +116 elo. Curiously SF 12 actually got a slightly higher average depth (.23 ply) despite only about 76% NPS.
I ran the same test on four threads, and got +142 -9 = 275 for 112 elo after 426 games, which is a smaller elo drop from 1 thread than would be expected normally. So it's probably fair to say it's about a hundred elo stronger than SF11 under more typical fast rapid time controls, but not 100 elo stronger than the strongest nonNNUE SF (July 31).
nnnnnnnn wrote: ↑Fri Sep 04, 2020 3:22 am
Stockfish 12 still cannot solve the win study [d]2R2K2/p3p3/3p4/p2P4/P7/4p3/6pp/6nk w - - 0 1 within a couple of minutes (4 cores, Mac, 8GB hash via brew). Even after the introductory moves 1. Rc1 e2, Stockfish 12 did not seem to see the winning move 2 Rb1! after a while.
I wish these engines were tuned better for the analysis of studies.