At what depth that is do more harm than good to Komodo or Stockfish?

[lkaufman]

We all know that Stockfish is much faster than Komodo Dragon, and therefore reaches a higher depth when paired at Bullets versus Komodo Dragon 2.5, but if a match was set to play at at pre determine depth lets say depth = 45 for both engines what engines would play better

Sometimes I wonder if you actually try to think about the gazillion questions you post.

If you set a fixed depth, you negate engine speed and time control. If it doesn't matter how long it takes to reach depth 45, it would even be best to omit any pruning, so the engine doesn't ever miss a tactic. In that case, the engines see the exact same positions to the exact same depth. So, the strongest one will be the one with the best evaluation. If you don't omit pruning, the engine which prunes the least has the ability to see more tactical shots, which would add strength to that engine.

When setting a fixed depth, you're not testing a chess engine, but pruning and evaluations.

I do not think the poster meant to test strength of the engine.
I see no reason to be against testing at fixed depth or with different time control or different depth for engines.

Mr. Uri thanks for making clear what I meant, I also wanted to know if after top engines reaches certain Depth let say above 40 if it benefit the outcome or it makes poor assumption of thinking that the opponent will always make the best move, the question of what this is equivalent to in terms of other programs, e.g. a null-mover with "standard" extensions, but what if it does NOT make the best move beyond depth 40 ?

Looking deeper is always beneficial against an opponent of the same class; if opponent looks 40 plies deep, we will score better looking 43 plies deep rather than 42. But against a vastly weaker opponent, say one looking 12 plies deep, it doesn't seem to help to outsearch him by more than about a dozen plies, so going from 24 to 36 or 48 plies won't help, even seems to hurt. But that's not measurable in standard chess, either version will score 100% (to nearest percentage point). We can only test that by handicap play. I have tested it in normal play with no change in win percentage (like 99.7% or so), but the statistical margin of error is too large even for thousands of games that way. With knight odds peak results on one thread for Dragon 2.5 seem to be at depth 23, probably would be depth 22 on four threads.

I noticed after 7 games at depth =40 between Dragon 2.5 vs Stockfish 14 that even if the score is even so far at 3.5, at Depth = 40 that Dragon 2.5 take 3 minutes per move to reach depth 40 while Stockfish 14 only needs 1 minute, therefore, when playing Blitz or Rapid chess Stockfish 14 is the King, but for Correspondence players Dragon 2.5 might be of a greatest help since it plays more like Top GM 's and given plenty of time to analyze a position and with the help of the correspondence World champion, well the choice of engine might be to have both analyzing the same position up to depth 80 or 100 if possible and at the end the human will choose the best move which might not be either Dragon 2.5 nor Stockfish 14 move, but certainly both engines can help the world correspondence champion determine which is the candidate move. also make me wonder at what depth will chess be solved with perfect play from both sides like checkers https://www.chess.com/news/view/compute ... chess-next

It may well be that chess is already "solved" in a weak sense, in that perhaps Dragon 2.5 or Stockfish on big hardware taking a day or more per move with no opening book might already be unbeatable no matter who or what is on the other side. But there is no practical way to test this, we'd need to play thousands of such games, and who would be the player or players on the other side? That doesn't mean it plays all positions perfectly, just that the draw margin in chess is large enough that the engine should never lose if allowed to choose its own openings. We can get an idea if this is true if several top level correspondence players play a hundred games or more each in the next year without losing a single game.

[Chessqueen]

We all know that Stockfish is much faster than Komodo Dragon, and therefore reaches a higher depth when paired at Bullets versus Komodo Dragon 2.5, but if a match was set to play at at pre determine depth lets say depth = 45 for both engines what engines would play better

Sometimes I wonder if you actually try to think about the gazillion questions you post.

If you set a fixed depth, you negate engine speed and time control. If it doesn't matter how long it takes to reach depth 45, it would even be best to omit any pruning, so the engine doesn't ever miss a tactic. In that case, the engines see the exact same positions to the exact same depth. So, the strongest one will be the one with the best evaluation. If you don't omit pruning, the engine which prunes the least has the ability to see more tactical shots, which would add strength to that engine.

When setting a fixed depth, you're not testing a chess engine, but pruning and evaluations.

I do not think the poster meant to test strength of the engine.
I see no reason to be against testing at fixed depth or with different time control or different depth for engines.

Mr. Uri thanks for making clear what I meant, I also wanted to know if after top engines reaches certain Depth let say above 40 if it benefit the outcome or it makes poor assumption of thinking that the opponent will always make the best move, the question of what this is equivalent to in terms of other programs, e.g. a null-mover with "standard" extensions, but what if it does NOT make the best move beyond depth 40 ?

Looking deeper is always beneficial against an opponent of the same class; if opponent looks 40 plies deep, we will score better looking 43 plies deep rather than 42. But against a vastly weaker opponent, say one looking 12 plies deep, it doesn't seem to help to outsearch him by more than about a dozen plies, so going from 24 to 36 or 48 plies won't help, even seems to hurt. But that's not measurable in standard chess, either version will score 100% (to nearest percentage point). We can only test that by handicap play. I have tested it in normal play with no change in win percentage (like 99.7% or so), but the statistical margin of error is too large even for thousands of games that way. With knight odds peak results on one thread for Dragon 2.5 seem to be at depth 23, probably would be depth 22 on four threads.

I noticed after 7 games at depth =40 between Dragon 2.5 vs Stockfish 14 that even if the score is even so far at 3.5, at Depth = 40 that Dragon 2.5 take 3 minutes per move to reach depth 40 while Stockfish 14 only needs 1 minute, therefore, when playing Blitz or Rapid chess Stockfish 14 is the King, but for Correspondence players Dragon 2.5 might be of a greatest help since it plays more like Top GM 's and given plenty of time to analyze a position and with the help of the correspondence World champion, well the choice of engine might be to have both analyzing the same position up to depth 80 or 100 if possible and at the end the human will choose the best move which might not be either Dragon 2.5 nor Stockfish 14 move, but certainly both engines can help the world correspondence champion determine which is the candidate move. also make me wonder at what depth will chess be solved with perfect play from both sides like checkers https://www.chess.com/news/view/compute ... chess-next

It may well be that chess is already "solved" in a weak sense, in that perhaps Dragon 2.5 or Stockfish on big hardware taking a day or more per move with no opening book might already be unbeatable no matter who or what is on the other side. But there is no practical way to test this, we'd need to play thousands of such games, and who would be the player or players on the other side? That doesn't mean it plays all positions perfectly, just that the draw margin in chess is large enough that the engine should never lose if allowed to choose its own openings. We can get an idea if this is true if several top level correspondence players play a hundred games or more each in the next year without losing a single game.

I went back to see each game played between Stock14 vs Dragon 2.5 every single one ended in a draw at depth 40, and to tell you the truth I was not happy to see so many draws. I know that I only let them play 10 games at depth 40, I know that it was a coincidence that all of the 10 games ended in a draw. It would be at least another 20 years before personal computer are fast enough for the top engine at that time to reach depth 100 in standard time control, and 99% of the games to end up in draw even if 1000 games are played, but we are not there yet. Checkers became not to much of interest when it was solved, and lots of people lost interest in playing checkers after that, hope that Chess will not become like checkers in the another 10 or 20 years, but eventually chess will be solved too, and as far as playing versus the top human Carlsen, it was solved more than 18 years ago since any of the top 5 engines will never lose a single game against him. Somebody should let the current Checkers champion Alexander Moiseyev to play against Chinook, but after Chinook solved checkers no sponsor would dare to pit a human against Chinook, just like no sponsor will waste money in sponsoring Magnus Carlsen versus any of the top 5 chess engines.