White to move |
The poster wrote that in the above position Stockfish 6 failed to find the best move after 18 hours of searching while Houdini 4 found it after a few seconds! I plugged the position into my program and after about 3 minutes Stockfish 6 (using 2 cores) wanted to play 1.Kf1, evaluating the position at 0.00. Komodo 8 liked 1.Rf8 with the same evaluation.
I have Houdini 2 and it took only a few seconds to find the correct 1.Bf8+, evaluated at 4.39. Houdini 1.5 also found 1.Bf8+ in a few seconds! I was curious to see what happened when I showed Stockfish 6 the correct 1.Bf8+. I did not give it a lot of time but just scrolled through the moves and it did not realize white was winning until I got to move 5 which was the same for Komodo 8. Further discussion can be found on the Rybka Forum HERE.
I checked several other engines (Black Mamba, Critter 1.6, Fritz 11, Gull 3 and Deep Rybka 4). Curiously, after about 20 seconds Critter 1.6 preferred 1.Bf8+ but evaluated the position at 0.00. But when I actually made the move, the evaluation immediately jumped to 5.00! The other engines did not find 1.Bf8+ and even after making the move on the board they didn't change the 0.00 evaluation.
This is an important lesson! If you are using Stockfish or Komodo to analyze, you should have Houdini running as a second engine!
29/70 00:59 87 074k 1 464k +6,06 1.Gd6-f8+ Kh6-h7 2.Wg8-g7+ Kh7-h8 3.Gf8-e7 Hh4-e1 4.Ge7-f6 He1xe2+ 5.Kg2-h1 He2-c4 6.Gf6-b2 h5-h4 7.Gb2-a1 h4-h3 8.Ga1-f6 Hc4-e6 9.Gf6-d4 He6-c4 10.Gd4-a1 h3-h2 11.Ga1-f6 Hc4-e6 12.Gf6-d4 He6-h6 13.Wg7-g6+ Kh8-h7 14.Wg6xh6+ Kh7xh6 15.Kh1xh2 Kh6-g5 16.Kh2-h3 Kg5-f5 17.Kh3-h4 Kf5-e6 18.Kh4-g4 Ke6-d5 19.Gd4-g1 Kd5-e5 20.Gg1-f2 Ke5-e6 21.Kg4xf4 Ke6-f6 22.Kf4-g4 Kf6-f7 23.Gf2-g3 Kf7-e6 24.Gg3-f4 Ke6-f6 25.Gf4-d6 Kf6-g6 26.f3-f4 Kg6-f6 27.Kg4-h3 Kf6-f5 28.Kh3-h4 Kf5-g6 29.Kh4-g3 Kg6-f5 30.Kg3-h3 Kf5-e4 31.Kh3-g4 Ke4-d4 32.f4-f5 Kd4-d5 33.Kg4-h4 Kd5-e4 34.f5-f6 Ke4-d5 35.f6-f7
ReplyDeleteFEN: 6R1/8/2pB3k/2P4p/5p1q/5P2/4P1K1/8 w - - 0 1
Houdini_3_AQ_x64_IDeA: