lichess.org
Donate

Kasparov is too deep for Stockfish

Whether the engine sees the move depends on the stockfish version. The server analysis did not see 27.h5; maybe someone has an outdated version of fishnet running on their server? I let my phone run to 31 depth and it didn't see it yet, even with "SF 14+ NNUE" showing on the lichess analysis page. Clearly newer versions are better at searching for this move.
@PieceDancer said in #9:
> I analysed it here, Stockfish 15.1, 1 core, 256 MB of hash, 7 seconds per move; move 27.h5 Stockfish does not see it and gives draw. 8 seconds per move Stockfish sees it, and says white wins.

Maybe it is your hashtable size, but most probably it is the time limit of 7 seconds per move. In my laptop (HP with AMD Ryzen 5 5500U 2.10 GHz, 8GB RAM, which is a very decent system), Stockfish 15.1 detects immediately the move, and all that followed, in "Infinite Analysis" mode of Fritz 14, 1 core, 1024 MB hash size. To be decent to the engine, one must use at least 4 cores, but this move, it can detect it with just one.

Other engines, almost equal in strength to Stockfish perform the same: Koivisto 8, Komodo 13, Slowchess Blitz 2.9, all detect the move at once. Deep Fritz 14, rated at 3064 by http://computerchess.org.uk/, detect 27.h4 at once, but for a few seconds it gives a rating of +0.32, before deciding it is a winning move. In this example we can see the real usefulness of extremely competent engines: given time, engines rated around 3000 (Deep Fritz 14, Rybka 2.3.2 etc) will "see" everything - they just require more time. Engines like Stockfish 15, Koivisto 8, Berserk etc, will complete their analysis much faster.
@phobbs5 said in #11:
> Whether the engine sees the move depends on the stockfish version. The server analysis did not see 27.h5; maybe someone has an outdated version of fishnet running on their server? I let my phone run to 31 depth and it didn't see it yet, even with "SF 14+ NNUE" showing on the lichess analysis page. Clearly newer versions are better at searching for this move.

Server analysis is just for quick and dirty job, and it is significantly less competent than analyzing with a pc, or laptop. In my laptop Stockfish 15 always reaches depth of 44, older Deep Fritz 14 easily reaches 27, and Rybka 2.3.2 reaches 32 most of the times (these two engiens are more than a decade old).

I do not expect that someone publishes an article about a game between Kasparov and Kramnik (mind you, Kramnik played extremely well in this game, except for one blunder, in a position where I guess everyone would blunder), and base his/her analysis on server engines, or even smartphone guis. It is one thing if Stockfish has difficulty with a move, when it analyses through a server, where resources are most of the times limited, and quite another when one claims publicly that Stockfish does not "appreciate the aesthetics" of the move Kasparov played.
How dare you critize Rb8 line in the Sveshnikov! Even Magnus Carlsen prefers it.
Idk. You guys need to have powerful hardware.
My Stockfish 16 dev version/stockfish 16 final, with a core i9 13900k CPU, saw 1 h5 instantly (in like 1/20th of a second), evaluation started at +3.13 and just kept growing, reaching +4.46 at depth 35.
I put 1 core on purpose, because the position was not so complicated.
@phobbs5 said in #11:
> Whether the engine sees the move depends on the stockfish version. The server analysis did not see 27.h5; maybe someone has an outdated version of fishnet running on their server? I let my phone run to 31 depth and it didn't see it yet, even with "SF 14+ NNUE" showing on the lichess analysis page. Clearly newer versions are better at searching for this move.

Server analysis is not good. I heard chess.com is even worse. Probably the hash size is too small. Not version of Stockfish.
It seems like we should open a bug against the fishnet repo to fix the hash size problem.