Hello there)
As usual, I am pleased to announce,I managed to run another new competitions...A new Duel match: NNUE vs NON-NNUE
A few quick notes,For the current Duel, I picked the strongest enginesFor example, as NON-NNUE the winner of CS 3600+As NNUE, Stockfish 011221, the winner of NNUE X
Note also that, (about NNUE vs NON-NNUE Duel)Running under these conditions is not much fair...!You may know, normally I prefer NNUE vs NNUEBut anyhow, I organized this tour.. just for comparing!Because I think that such testing is missing, right ?)Plus many chess friends are talking and talking thatThere is a lot of Elo difference such as 300 even 500 Elo
To be more clear, Definitely for a better conclusion we need more games,But anyhow, the current new small data shows us that,NNUE is incredibly superior strong, really impressive!NNUE plays less blunder moves and better, stronger games!For this reason, especially with Blitz time controls,The Evalfies play a BIG role..where NON-NNUE suffers...
But this is also true, and I have to admit that,The difference is not 300 Elo, 500 Elo etc...Perhaps with very weak openings, 1 core, Ponder OFF etcOr via ultra fast time controls we can expect more ?Perhaps testing the engines at TC 20 sec/game ?) Sure under these conditions anything can happen...Frankly, I can't say that it is a real chess, sorry...
On other hand, in case of running serious tournaments,E.g let's say with slower TC such 40/120, MP, Ponder ON:Then I expect NNUE vs NON-NNUE to be less than 50 Elo Sure in case of running a tour with strong short openings!But not 100+ Elo difference for sure...it will be just a dream)Btw, as alternative...not so bad idea to be compared:SSDF
So in this Duel, it is not a BIG surprise that,The played results are in favor for NNUE (EvalFile)Because NNUE gains in openings, middle-game, endgame etcWhere NON-NNUE engine is playing without any assistance...That's why I say...is it much fair...?! sure NO...
Meanwhile, we should not forget to mention aboutThe real power of strongest deep opening theory as well!Actually this is known issue since a long time, once more:Just in case of NON-NNUE Blitz matches: Best/strongest deep books are 100+ Elo stronger than short booksE.g In case of via ultra fast time controls may appear even 200+ EloJust saying !) and hope helps...
And here are NNUE vs NON-NNUE's Duel results:And I hope my current data will be helpful...
Blitz 3m+2s - Elo Difference: 152
1 Stockfish 011221 +52/-1/=71 70.56% 87.5/1242 Cfish 030820 +1/-52/=71 29.44% 36.5/124
*Blitz games are played with 6 Core/Game, 512 MB Hash--------------------------------------------------------------------------------------------
Rapid 15m+5s - Elo Difference: 87
1 Stockfish 011221 +31/-0/=95 62.30% 78.5/1262 Cfish 030820 +0/-31/=95 37.70% 47.5/126
*Rapid games are played with 4 Core/Game, 1024 MB Hash
GAMES
Best,Sedat
As usual, I am pleased to announce,I managed to run another new competitions...A new Duel match: NNUE vs NON-NNUE
A few quick notes,For the current Duel, I picked the strongest enginesFor example, as NON-NNUE the winner of CS 3600+As NNUE, Stockfish 011221, the winner of NNUE X
Note also that, (about NNUE vs NON-NNUE Duel)Running under these conditions is not much fair...!You may know, normally I prefer NNUE vs NNUEBut anyhow, I organized this tour.. just for comparing!Because I think that such testing is missing, right ?)Plus many chess friends are talking and talking thatThere is a lot of Elo difference such as 300 even 500 Elo
To be more clear, Definitely for a better conclusion we need more games,But anyhow, the current new small data shows us that,NNUE is incredibly superior strong, really impressive!NNUE plays less blunder moves and better, stronger games!For this reason, especially with Blitz time controls,The Evalfies play a BIG role..where NON-NNUE suffers...
But this is also true, and I have to admit that,The difference is not 300 Elo, 500 Elo etc...Perhaps with very weak openings, 1 core, Ponder OFF etcOr via ultra fast time controls we can expect more ?Perhaps testing the engines at TC 20 sec/game ?) Sure under these conditions anything can happen...Frankly, I can't say that it is a real chess, sorry...
On other hand, in case of running serious tournaments,E.g let's say with slower TC such 40/120, MP, Ponder ON:Then I expect NNUE vs NON-NNUE to be less than 50 Elo Sure in case of running a tour with strong short openings!But not 100+ Elo difference for sure...it will be just a dream)Btw, as alternative...not so bad idea to be compared:SSDF
So in this Duel, it is not a BIG surprise that,The played results are in favor for NNUE (EvalFile)Because NNUE gains in openings, middle-game, endgame etcWhere NON-NNUE engine is playing without any assistance...That's why I say...is it much fair...?! sure NO...
Meanwhile, we should not forget to mention aboutThe real power of strongest deep opening theory as well!Actually this is known issue since a long time, once more:Just in case of NON-NNUE Blitz matches: Best/strongest deep books are 100+ Elo stronger than short booksE.g In case of via ultra fast time controls may appear even 200+ EloJust saying !) and hope helps...
And here are NNUE vs NON-NNUE's Duel results:And I hope my current data will be helpful...
Blitz 3m+2s - Elo Difference: 152
1 Stockfish 011221 +52/-1/=71 70.56% 87.5/1242 Cfish 030820 +1/-52/=71 29.44% 36.5/124
*Blitz games are played with 6 Core/Game, 512 MB Hash--------------------------------------------------------------------------------------------
Rapid 15m+5s - Elo Difference: 87
1 Stockfish 011221 +31/-0/=95 62.30% 78.5/1262 Cfish 030820 +0/-31/=95 37.70% 47.5/126
*Rapid games are played with 4 Core/Game, 1024 MB Hash
GAMES
Best,Sedat