NNUE at LTC

[mwyoung]

i see from mwyoung tournaments (TC 10min,30mins,32 threads) that nnue vs sf is close to =

Here is the current 30m + 30s results. Clearly not as impressive as with short TC and lower threads.

DESKTOP-CORSAIR, Rapid 30.0min+30.0sec 0


1 Stockfish 170720 64 POPCNT +0/=22/-0 50.00% 11.0/22 121.00
2 SF+NNUE PO 290720 x64 popc +0/=22/-0 50.00% 11.0/22 121.00

[carldaman]

You are not specifying which net you're using.

[mwyoung]

You are not specifying which net you're using.

http://talkchess.com/forum3/viewtopic.php?f=6&t=74630

Yes I did, just not here.

[carldaman]

OK, thanks, interesting. Wouldn't that be a huge letdown if NNUE did not scale well at all!

[Gino Figlio]

NNUE and regular stockfish are the same in correspondence time control or analysis

I find them completely different, worlds apart. You can always guide the position to places where NNUE outperforms regular Stockfish (just give bonuses for positions where they disagree, as the disagreement grows you reach positions where regular Stockfish's eval isn't good at all.)

I define cc time control as an overnight analysis of a position, 12 hours, reaching depths of 70-90. Top move is always the same. I am not able to show positions of ongoing games but you can test using these conditions to confirm my statement (need at least two similar computers running simultaneously)

[Zenmastur]

NNUE and regular stockfish are the same in correspondence time control or analysis

I find them completely different, worlds apart. You can always guide the position to places where NNUE outperforms regular Stockfish (just give bonuses for positions where they disagree, as the disagreement grows you reach positions where regular Stockfish's eval isn't good at all.)

I define cc time control as an overnight analysis of a position, 12 hours, reaching depths of 70-90. Top move is always the same. I am not able to show positions of ongoing games but you can test using these conditions to confirm my statement (need at least two similar computers running simultaneously)

First, I have done this myself, and it's my experience that the top moves aren't always the same. Second, you're talking about unassisted analysis. Assisted analysis can yield different lines of play and radically different evaluations.

[Gino Figlio]

I am not sure you are disagreeing with my statement.
Top move (not moves) is always the same.
Unassisted analysis takes the human skill out of the equation, to make the comparison fair

[Ovyron]

Unassisted analysis takes the human skill out of the equation, to make the comparison fair

Okay, but then correspondence time controls are irrelevant. You can get some 72 hours unassisted analysis in 1 hour of interactive analysis so the only conclusion of letting the computer run overnight is "you're wasting your time", traditional Stockfish or not.

[Leto]

Unassisted analysis takes the human skill out of the equation, to make the comparison fair

Okay, but then correspondence time controls are irrelevant. You can get some 72 hours unassisted analysis in 1 hour of interactive analysis so the only conclusion of letting the computer run overnight is "you're wasting your time", traditional Stockfish or not.

As an active correspondence player I agree with Ovyron, assisted analysis if done by a competent correspondence player is of much higher quality than any unassisted analysis, unless the unassisted analysis is done by an engine that plays almost perfect chess. I believe no engine exists today that plays almost perfect chess, even these new NNUE nets.

[Gino Figlio]

Agreed. What I said was that Stockfish NNUE at correspondence time control does not bring anything more than what standard Stockfish provides