2 Bishops, 9 Pawns: only for big machines!

[Laskos]

Pierre Nolot sent me this position by Mark Dvoretsky, calling it as one of his favorites!

Black to move, only Kf6! wins.

How many hours (days...) on your machine?
Houdini? or other?
Do endgame tablebases help?

[d]8/4p1kp/3p2p1/2pP2b1/B1P1P1P1/K7/8/8 b - - bm Kg7f6!; (Dvor) 124

I will be surprised if houdini is the fastest in some endgame test position.

Endgames are not the strong part of houdini.

I remember that stockfish was clearly stronger than houdini in winning against tablebases test and I think that even with slightly more pieces stockfish is stronger.

Correct, these were the results for 3-4-5 TB wins in 30-40 moves.

On 360 hard 3-4-5 TB mates in 30-40 moves, without EGTB at 3,000ms/move the results are (number of wins on won TB positions of the engines without EGTB against Houdini 3 enabled with 3-4-5 Nalimovs)

Code: Select all

 
Stockfish 2.3.1  80% (+42) 
Komodo 5         70%  (+4) 
Gaviota 0.86     66%  (+9) 
Hiarcs 14        64% (+26) 
Hannibal 1.3     61%  (+8) 
Shredder 12      58% (+20) 
Critter 1.6      53% (+42) 
Houdini 3        46% (+34) 
Rybka 4.1        39% (+32) 
Junior 13        32% (+17) 

The numbers in parenthesis are improvements from 100ms/move. Stockfish performs the best and improves the most on TB positions, Rybka 4.1 and Junior 13 are the worst in finding endgame TB wins.

So yes, Stockfish seems stronger than Houdini in endgames, at least in those with few pieces.

Unfortunately I do not know about endgame rating list when you start only from positions when white and black have not more than one single piece.

[ernest]

you definitely don't need big machine here with Quazar, using one core only anyhow,

Indeed, Peter, good find with Quazar!

In my tests, ...Kf6 immediately appears in the PV, and the eval climbs (sometimes fast, sometimes not so fast). Good!

But if you try multiPV, MPV=2, well that's another story, and here ...h5 is back !!!

herebelow, after depth 47:
Analysis by Quazar 0.4 w32:

1. -+ (-6.64): 1...h5 2.gxh5 gxh5 3.Kb3 Kf6 4.Kc3 Ke5 5.Kd3 Kf4 6.Ke2 Kg3 7.Kf1 Be3 8.Ke2 Bd4 9.Be8 h4 10.Bh5 h3 11.Bf3 Bf6 12.e5 dxe5 13.d6 exd6 14.Bc6 Kf4 15.Kd3 h2 16.Be4 Bh4 17.Bb7 Bf2 18.Bd5 Be3 19.Be4 Bd4 20.Bc6 Kg4 21.Be4
2. -+ (-4.58): 1...Kf6 2.Bd1 Ke5 3.Ka4 Bf6 4.Kb3 Kxe4 5.Ka3 Kd3 6.Kb3 Ke3 7.Kc2 Bg7 8.Kb3 Kd2 9.Bf3 Bf6 10.Ka4 Kc3 11.Kb5 Kd4 12.Bd1 Ke4 13.Ba4 Kf4 14.Bc2 Kg5 15.Bd1 h5 16.gxh5 gxh5 17.Be2 Be5 18.Bf1 Kf4 19.Kc6 h4

So something seems strange with Quazar...

[peter]

[But if you try multiPV, MPV=2, well that's another story, and here ...h5 is back !!!

Oops, you'r right, Ernest:


8/4p1kp/3p2p1/2pP2b1/B1P1P1P1/K7/8/8 b - -

Engine: Quazar 0.4 x64 (4096 MB)
von Dmitry Morozov

55 94:22 -6.65 1...h5 2.gxh5 gxh5 3.Kb3 Kf6 4.Kc3 Ke5
5.Kd3 Kf4 6.Ke2 Kg3 7.Kf1 Le3 8.Ke2 Lf2
9.Le8 h4 10.Lh5 Ld4 11.Lf3 Lf6
12.Lh5 h3 13.Lf3 Ld4 14.e5 dxe5 (5.399.365.741) 953

55 94:22 -4.37 1...Kf6 2.Ld1 Lf4 3.Kb2 Kg5 4.Kc2 h5
5.gxh5 gxh5 6.Kd3 h4 7.Ke2 h3 8.Kf2 Kf6
9.Kf3 h2 10.Kg2 Ke5 11.La4 Kxe4
12.Kh1 Ke5 13.Ld7 Lg3 14.Lb5 Kf4 (5.399.365.741) 953


Seems the second primary variant cost' too much computing power of the single core. Maybe the difference to non primary variants' pruning is the reason for not getting the point of 1...Kf6 in this way,