syzygy wrote: ↑Sat Oct 10, 2020 1:55 pm
Why not!?
Why would you be allowed to tune your non-NNUE eval by playing games against Stockfish if you're not allowed to tune your NNUE eval with games played by (or rather: positions generated by) Stockfish.
Why not? Well either incompetence or malice. I'm not sold on which one it is, but I'm certain its at least one. Potentially both.
Good work on your CFish NNUE implementation, by the way. I was blown away that I could wrap Ethereal around it with an hour of work, and the only "copied" part other than the NNUE files itself was Stockfish's misc.c and misc.h helpers.
I'm still not entirely convinced that the accumulator is 64byte aligned in all cases, but it seems to work.
Talkchess is dead without moderation. If you want my attention, contact me via andrew@grantnet.us
Raphexon wrote: ↑Sat Oct 10, 2020 2:14 pm
Eth-NNUE does seem to be a lot slower than Cfish and SF.
Only 300 knps (from start) on 1 core is really low. SFnnue always did over double.
I mean, its the same code. So, its an issue with your compilation probably.
Talkchess is dead without moderation. If you want my attention, contact me via andrew@grantnet.us