22

Letting Stockfish (14+ NNUE) run for several minutes with lichess' analysis tool on the initial position of a standard chess board lets it reach depths beyond 30 moves, I suppose letting it run for many hours or days would result in an even deeper evaluation.

What is the deepest evaluation ever done of this kind?

2080
  • 916
  • 8
  • 23

2 Answers2

22

In the chessbase cloud, this depth 79 evaluation is displayed:

enter image description here

Which means someone used Chessbase to get their Stockfish 12 engine to calculate to depth 79, and they had an internet connection at the time so it was able to be stored on the cloud. However, there are some caveats:

  1. We don't know the engine settings that this person was using -- although I doubt this matters much though for the calculation time required, relatively speaking.
  2. There could have been a deeper search done by someone, who either wasn't connected to the cloud at the time, or maybe wasn't even using Chessbase at all.
  3. Chessbase might have an even higher depth evaluation stored in its cloud for Stockfish, but they just choose to show this depth 79 one (I don't know how this works).

But all in all, depth 79 seems like a good ballpark at the very least.

Also as @TonyK mentioned, Stockfish depth is in plies (so depth 79 refers to roughly 40 moves). Also Stockfish does very aggressive pruning. An exhaustive depth 79 search would involve something on the order of 30^79 nodes.

Inertial Ignorance
  • 19,818
  • 23
  • 70
16

Don't for one minute imagine that Stockfish has analysed all lines out to 30 plies (yes, that's plies, not moves). It has analysed some lines up to 30 plies, but most lines have been discarded well below that depth as being unproductive. Occasionally it will erroneously discard productive lines, but Stockfish being Stockfish, this doesn't happen very often.

TonyK
  • 905
  • 4
  • 15
  • 4
    Good point mentioning the 10^30 tree is way too big for a fully complete search. But this should probably be a comment. – Inertial Ignorance Nov 01 '22 at 06:36
  • Would be interesting to have a position where only few moves are possible for both sides all the time (and many can be pruned due to definite loss). Think of a fast tablebase "in situ". – Hauke Reddmann Nov 01 '22 at 10:03
  • 3
    @InertialIgnorance - Just to be picky, when we're talking on orders of magnitude as big as x^30... x being off by a factor of 3 is basically negligible – ScottishTapWater Nov 01 '22 at 10:52
  • 13
    @ScottishTapWater 30^30 / 10^30 is about 200 trillion though. – Inertial Ignorance Nov 01 '22 at 11:06
  • Ahh yeah no, you're right... I was thrown off by it being 30^30 not 30 x 10^30... This actually results in being off by a factor of 3^30 so not so negligible afterall – ScottishTapWater Nov 01 '22 at 11:35
  • And it looks like it incorrectly discarded 1. e4 as well. 1. d4 can't possibly be better :) – JK. Nov 04 '22 at 03:32