Thanks for the position Mc Mad!
It seemed a good opportunity for the first try-out of Stockfish 1.4 on my machine. Compared to Rybka's analysis, it is strange, Stockfish does not really consider exd4, only as the third move in the move list, but gxh5 is also not a certain draw when I look at the eval, so Rybka must be seeing something different here? Maybe deeper analysis is necessary with Rybka.
Freshly installed Stockfish 1.4 is fast in discarding Bxh5 as the best move, at one point Stockfish now almost reaches double the plydepth that Ancalagon reaches in the same time

At least for build 189 in this position. This does not say everything, but it must be said that Jim Ablett's build of 32-bit Stockfish 1.4 seems certainly a bit faster in nodes per second than the Stockfish 1.3 32-bit. I have not made any measurements but compared to Ancalogon Stockfish 1.4 seems a faster compile. It could be partly because of better, faster code from Marco and Joona that a higher nodes per second now is reached. But Jim's executable is also a bit smaller now, only 167 kB where the old one was 276 kB. The 64-bit build Stockfish 1.4 is more than 5 times as large as the 32-bit! Is that because of other libraries needed or something like that? I would imagine the optimization does not differ a lot for 32-bit and 64-bit, things like loop unrolling etc. should be the same roughly...
[d]rn1q1rk1/ppp2nbp/3p1pp1/4p2P/3PP1b1/2NBBN2/PPP1QPP1/2KR3R b - -
Engine: Stockfish 1.4 JA (256 MB)
by Tord Romstad, Marco Costalba, Joona Kii
2.00 0:00 -1.03 1...Bxh5 2.Kb1 (4.538) 11
3.00 0:00 -0.46 1...Bxh5 2.Kb1 Nc6 (4.943) 12
4.00 0:00 -0.58 1...Bxh5 2.dxe5 Nxe5 3.Kb1 Nxd3
4.Qxd3 (7.483) 17
5.00 0:00 -0.34 1...Bxh5 2.Kb1 Nc6 3.Nb5 a5 (11.446) 27
6.00 0:00 -0.32 1...Bxh5 2.dxe5 fxe5 3.Kb1 Nc6 4.b3 (16.515) 37
7.00 0:00 -0.22 1...Bxh5 2.dxe5 fxe5 3.Kb1 Nc6 4.b3 Nd4 (25.271) 55
8.00 0:00 -0.28 1...Bxh5 2.d5 Nd7 3.Kb1 Ng5 4.b3 Nc5
5.Bxc5 Nxf3 (44.383) 88
9.00 0:00 -0.26 1...Bxh5 2.d5 Nd7 3.Kb1 Ng5 4.b3 a6
5.Rdg1 b5 (76.886) 140
10.00 0:00 -0.06 1...Bxh5 (216.765) 261
11.00 0:01 -0.22 1...Bxh5 2.Rdg1 exd4 3.Bxd4 c5 4.g4 cxd4
5.Nd5 Bh6+ 6.Kb1 Bxg4 7.Rxg4 Nc6 (576.793) 372
12.01 0:02 -0.22 1...Bxh5 2.Rdg1 exd4 3.Bxd4 c5 4.g4 cxd4
5.Nd5 Bh6+ 6.Kb1 Bxg4 7.Rxg4 Nc6
8.a3 (1.274.995) 436
13.01 0:06 -0.12 1...Bxh5 2.Rdg1 Bg4 3.d5 Nd7 4.Kb1 c6
5.Rh4 h5 6.b3 Nb6 7.Rgh1 cxd5 8.exd5 (2.920.387) 484
14.01 0:10 -0.32 1...Bxh5 2.dxe5 (5.412.109) 497
15.01 0:25 -0.62 1...Bxh5 2.dxe5 fxe5 3.Bc4 Bg4 4.Qd2 Nd7
5.Ng5 Bxd1 6.Qxd1 Nf6 7.Ne6 Qd7
8.Nxf8 Kxf8 9.Kb1 c6 (12.825.212) 512
15.02 0:38 -0.34 1...gxh5 2.d5 c6 3.Kb1 Nd7 4.b3 a6
5.Qe1 Nc5 6.Bc4 cxd5 7.Bxd5 Rc8
8.Na4 Nxa4 9.bxa4 (19.838.444) 512
16.01 1:48 -0.44 1...gxh5 2.Qe1 Nc6 3.Be2 a6 4.d5 Ne7
5.Nd2 f5 6.f3 f4 7.Bf2 Bd7 8.Rxh5 c6
9.Bc4 (55.550.546) 511
17.01 2:38 -0.46 1...gxh5 2.Qe1 Nc6 3.Be2 a6 4.d5 Ne7
5.Nd2 c6 6.dxc6 bxc6 7.f3 Be6 8.Rxh5 Ng5
9.Kb1 d5 (80.784.205) 511
18.01 4:56 -0.40 1...gxh5 2.Qe1 Nc6 3.Be2 f5 4.dxe5 fxe4
5.Nxe4 Qe7 6.Nf6+ Bxf6 7.exf6 Qxf6
8.Kb1 a6 9.b3 b5 10.Qd2 Nfe5 11.Nxe5 Qxe5 (150.422.002) 508
best move: g6xh5 time: 5:19.969 min n/s: 508.763 nodes: 162.780.268
Some of the latest Ancalagon builds:
rn1q1rk1/ppp2nbp/3p1pp1/4p2P/3PP1b1/2NBBN2/PPP1QPP1/2KR3R b - -
Engine: Ancalagon 1.3 Weak Squares 180 Board Control middle game 50 - endgame 50 Build 189 (Athlon 2009 MHz, 256 MB)
by Romstad, Costalba, Kiiski, de Groot
2.00 0:00 -0.90 1...Bxh5 2.Kb1 Nc6 (8.911) 33
3.00 0:00 -0.74 1...Bxh5 2.Kb1 Nc6 3.Bc4 exd4 4.Bxd4 Nxd4
5.Rxd4 (51.713) 143
4.00 0:00 -0.74 1...Bxh5 2.Kb1 Nc6 3.Bc4 exd4 4.Bxd4 Nxd4
5.Rxd4 (69.569) 177
5.00 0:00 -0.88 1...Bxh5 2.Kb1 Nc6 3.dxe5 fxe5 4.Bc4 (112.053) 238
6.00 0:00 -0.62 1...Bxh5 2.Kb1 Nc6 3.dxe5 fxe5 4.Bc4 Nd4 (255.794) 333
7.01 0:03 -0.50 1...Bxh5 2.Rxh5 gxh5 3.Rh1 c6 4.Nd2 h4
5.Qh5 (1.634.579) 460
8.01 0:04 -1.00 1...Bxh5 2.Rxh5 (2.228.649) 491
9.01 4:30 -1.43 1...Bxh5 2.Rxh5 gxh5 3.Bc4 h6 4.Rh1 Qd7
5.Rxh5 Qg4 6.Rh2 Nc6 (126.637.434) 468
9.02 4:44 -1.07 1...gxh5 2.Bc4 Nc6 3.Qd2 Qe8 4.dxe5 fxe5
5.Qd5 Kh8 (133.272.359) 469
10.01 6:02 -0.94 1...gxh5 2.Bc4 exd4 3.Bxd4 Nc6 4.Be3 f5
5.exf5 Qf6 6.Kb1 Rae8 7.Bb3 Qxf5 (170.703.220) 470
11.01 9:17 -1.07 1...gxh5 2.Qd2 Nc6 3.Bc4 f5 4.Bg5 Qe8
5.Bh6 f4 6.Bxg7 Kxg7 7.dxe5 Bxf3
8.gxf3 Qxe5 (262.875.785) 471
12.01 11:42 -1.21 1...gxh5 2.Qd2 Nc6 3.Bc4 f5 4.Bg5 Qd7
5.dxe5 fxe4 6.Nxe4 Ncxe5 7.Nxe5 Bxe5
8.Nf6+ Bxf6 9.Bxf6 Bxd1 10.Rxd1 (334.205.378) 476
13.01 21:20 -1.23 1...gxh5 2.Qd2 Nc6 3.Bc4 Na5 4.Be2 b6
5.Kb1 c6 6.Nh4 Bxe2 7.Qxe2 exd4
8.Rxd4 (611.630.780) 477
best move: g6xh5 time: 27:11.109 min n/s: 480.230 nodes: 783.210.236
Grrr, more than 4 minutes to find gxh5. But Ancalagon is more intended for long analysis times so it should not be a real surprise. The build-up of the plydepths is just very different. Stockfish prunes wherever possible and gets its strength from phenomenal plydepth. I think it is a bit like Rybka
rn1q1rk1/ppp2nbp/3p1pp1/4p2P/3PP1b1/2NBBN2/PPP1QPP1/2KR3R b - -
Engine: Ancalagon 1.3 WS180BC5050 Build 190 (256 MB)
by Romstad, Costalba, Kiiski, de Groot
2.00 0:00 -0.90 1...Bxh5 2.Kb1 Nc6 (8.911) 35
3.00 0:00 -0.74 1...Bxh5 2.Kb1 Nc6 3.Bc4 exd4 4.Bxd4 Nxd4
5.Rxd4 (52.288) 152
4.00 0:00 -0.74 1...Bxh5 2.Kb1 Nc6 3.Bc4 exd4 4.Bxd4 Nxd4
5.Rxd4 (73.693) 196
5.00 0:00 -0.88 1...Bxh5 2.Kb1 Nc6 3.dxe5 fxe5 4.Bc4 (118.583) 261
6.00 0:00 -0.62 1...Bxh5 2.Kb1 Nc6 3.dxe5 fxe5 4.Bc4 Nd4 (279.223) 364
7.01 0:04 -1.27 1...Bxh5 2.Rxh5 (1.952.643) 473
7.03 0:06 -1.26 1...gxh5 (2.870.856) 472
8.01 0:07 -1.27 1...gxh5 2.Kb1 Nc6 3.Bc4 exd4 4.Bxd4 Nxd4
5.Rxd4 Bh6 6.Nd5 (3.932.333) 492
8.02 0:08 -1.26 1...Bxh5 (4.057.784) 496
9.01 0:33 -1.52 1...Bxh5 2.dxe5 fxe5 3.Bc4 Qd7
4.Rxh5 gxh5 5.Ng5 Qg4 6.Bxf7+ Rxf7
7.Qc4 Qd7 8.Nxf7 Qxf7 (17.282.643) 516
9.02 0:38 -0.98 1...gxh5 2.Kb1 Nc6 3.Bc4 exd4 4.Bxd4 Nxd4
5.Rxd4 Bh6 6.Nd5 c6 (19.673.068) 509
10.01 2:05 -0.92 1...gxh5 2.Bc4 Nc6 3.Qf1 Na5 4.Kb1 c6
5.Be2 b5 6.Qe1 (60.775.163) 483
11.01 9:28 -1.09 1...gxh5 2.Qd2 Nc6 3.Bc4 Na5 4.Be2 a6
5.Kb1 b5 6.dxe5 fxe5 7.Nd5 (268.602.592) 472
12.01 17:57 -1.13 1...gxh5 2.Qd2 Nc6 3.dxe5 dxe5 4.Nh4 Nb4
5.Nf5 Nxd3+ 6.cxd3 Bxf5 7.exf5 Qd7
8.Rxh5 Rfd8 9.Qc2 Kf8 (508.498.953) 471
13.01 26:14 -1.15 1...gxh5 2.Qd2 Nc6 3.Bc4 f5 4.dxe5 fxe4
5.e6 exf3 6.exf7+ Kh8 7.g3 Qd7 8.Qd5 Ne5
9.Bb3 Qxf7 10.Qxb7 (744.852.418) 473
14.01 128:36 -1.35 1...gxh5 2.Qd2 (3.112.166.139) 403
The evaluation for 10...gxh5 seems to be sliding precariously away from a draw. It may be an illusion though, King Safety can give White a plus but it does not seem decisive if it is King Safety only, and King Safety scores are much larger in Ancalagon anyway.
Good testposition!
Eelco