Vayser-Overton, WC35/sf12 ICCF 2011

[jdart]

[D] 2b3r1/8/1kp1p1P1/2b1P2B/1pprPP2/7p/1P5Q/2KN4 b - - 0 38

Black played Ba6!! and White resigned. This is better than Rxe4.

The main line is Qxh3 (other moves are worse), Bb5! and Black will push the b/c pawns, with a winning attack. But it is a very complex position.

--Jon

[kgburcham]

good test position Jon.
[D] 2b3r1/8/1kp1p1P1/2b1P2B/1pprPP2/7p/1P5Q/2KN4 b - -

Houdini

25/76 2:16 -1.80 1...Rxe4 2.Qh1 Rxf4 3.Qxh3 Rd4 4.Kb1 Bd7 (2.034.088.778) 14933
25/76 3:21 -1.89++ 1...Ba6 (2.999.421.802) 14888
25/76 3:55 -2.09++ 1...Ba6 (3.489.806.103) 14844
25/76 5:24 -2.52++ 1...Ba6 (4.810.631.945) 14813
25/76 11:18 -3.68++ 1...Ba6 (10.164.250.377) 14971
25/83 34:17 -6.78++ 1...Ba6 (31.462.521.473) 15289
25/83 57:09 -7.45 1...Ba6 2.Qxh3 Ra8 3.g7 Bb5 4.Kb1 c3 5.bxc3 Bd3+ (52.687.013.799) 15364

[Houdini]

good test position Jon.
[D] 2b3r1/8/1kp1p1P1/2b1P2B/1pprPP2/7p/1P5Q/2KN4 b - -

Houdini

25/76 2:16 -1.80 1...Rxe4 2.Qh1 Rxf4 3.Qxh3 Rd4 4.Kb1 Bd7 (2.034.088.778) 14933
25/76 3:21 -1.89++ 1...Ba6 (2.999.421.802) 14888
25/76 3:55 -2.09++ 1...Ba6 (3.489.806.103) 14844
25/76 5:24 -2.52++ 1...Ba6 (4.810.631.945) 14813
25/76 11:18 -3.68++ 1...Ba6 (10.164.250.377) 14971
25/83 34:17 -6.78++ 1...Ba6 (31.462.521.473) 15289
25/83 57:09 -7.45 1...Ba6 2.Qxh3 Ra8 3.g7 Bb5 4.Kb1 c3 5.bxc3 Bd3+ (52.687.013.799) 15364

I agree that this is a very good and complex test position: Ba6, Ra8, Bb5 is a very deep maneuver against which White apparently is helpless.

Kim, this is another good example position for the "accelerate finding a PV after a fail high" trick that we discussed in another thread.
In your analysis the fail high is established in 3 minutes (3 billion nodes), but it takes 20 times more time (another 50 billion nodes) to find the PV.
Below the same with the current Houdini dev version, on slower hardware:

27/75 6:09 -1.89 1...Rxe4 2.Qg3 Rd4 3.Qxh3 Rxf4 4.Kb1 Rd4 5.Ne3 Re4 ... (2.276.715.952) 6165
27/75 9:44 -2.00++ 1...Ba6 (3.596.469.477) 6147
27/75 11:18 -2.20++ 1...Ba6 (4.168.383.088) 6140
24/75 11:49 -2.67++ 1...Ba6 (4.351.589.342) 6129
24/75 14:26 -3.94++ 1...Ba6 (5.291.249.062) 6104
24/90 40:16 -6.42 1...Ba6 2.f5 exf5 3.e6 fxe4 4.Be2 Rxg6 5.Qe5 Rg8 ... (14.799.768.838) 6123

The fail high is found after 3.5 billion nodes, the PV is available 11 billion nodes later, nearly 5 times faster than in your analysis.

[Jouni]

Where is the progress - Houdini 1.5a gets this faster

Analysis by Houdini 1.5a x64:

1...Rxe4 2.Qh1
-/+ (-0.88 --) Depth: 24/72 00:05:20 895mN, tb=1
1...Ba6
-/+ (-1.19 ++) Depth: 24/72 00:08:40 1469mN, tb=1

[kgburcham]

Where is the progress - Houdini 1.5a gets this faster
Analysis by Houdini 1.5a x64:
1...Rxe4 2.Qh1
-/+ (-0.88 --) Depth: 24/72 00:05:20 895mN, tb=1
1...Ba6
-/+ (-1.19 ++) Depth: 24/72 00:08:40 1469mN, tb=1

[D] 2b3r1/8/1kp1p1P1/2b1P2B/1pprPP2/7p/1P5Q/2KN4 b - -

Houdini 2.0c

24/57 0:18 -1.18++ 1...Ba6 (628.346.599) 34871
24/59 0:21 -1.37++ 1...Ba6 (753.005.524) 34846
24/59 0:32 -1.80++ 1...Ba6 (1.127.619.934) 35151
24/81 1:28 -2.96++ 1...Ba6 (3.237.259.651) 36512

slower clock
25/65 0:22 -1.31-- 1...Rxe4 2.Qh1 (678.479.532) 30251
25/65 0:46 -1.51++ 1...Ba6 (1.439.342.449) 31160
25/65 1:00 -1.70++ 1...Ba6 (1.904.012.486) 31304
25/65 1:33 -2.14++ 1...Ba6 (2.964.541.175) 31791
25/65 4:16 -3.30++ 1...Ba6 (8.475.407.948) 33103
25/84 10:18 -3.30 1...Ba6 2.Qxh3 Ra8 3.f5 exf5 4.Nc3 bxc3 (21.118.461.596) 34121 TB:10

[Houdini]

Where is the progress - Houdini 1.5a gets this faster

Analysis by Houdini 1.5a x64:

1...Rxe4 2.Qh1
-/+ (-0.88 --) Depth: 24/72 00:05:20 895mN, tb=1
1...Ba6
-/+ (-1.19 ++) Depth: 24/72 00:08:40 1469mN, tb=1

SMP makes this a lottery.
Here's my run with Houdini 1.5a (3 threads, 2048 MB hash), it only finds Ba6 at depth 26, meaning it's about 4 times slower than your run:

25/71 2:43 -0.91 1...Rxe4 2.Qh1 Rd4 3.Qxh3 Rxf4 4.Kb1 Bd7 5.Kc1 Kc7 ... (999.041.786) 6128
26/77 4:02 -0.98++ 1...Rxe4 (1.497.762.245) 6180
26/77 5:32 -1.14++ 1...Rxe4 (2.057.357.459) 6184
26/80 8:00 -1.14 1...Rxe4 2.Qh1 Rd4 3.Qxh3 Rxf4 4.Kc2 Rd4 5.Ne3 Re4 ... (2.979.469.513) 6204
26/80 13:17 -1.55++ 1...Ba6 (4.948.637.140) 6205
26/80 26:19 -2.72++ 1...Ba6 (9.851.151.438) 6238

Finding the PV will take ages...

[ernest]

SMP makes this a lottery.

So true..., and even with single core lottery comes from the choice of hash size!

[marcelk]

[D] 2b3r1/8/1kp1p1P1/2b1P2B/1pprPP2/7p/1P5Q/2KN4 b - - 0 38

Black played Ba6!! and White resigned. This is better than Rxe4.

The main line is Qxh3 (other moves are worse), Bb5! and Black will push the b/c pawns, with a winning attack. But it is a very complex position.

--Jon

Rookie gets its fail high on a different one, Rg7. I'm not sure what to think of the rook shuffle Rg7 ... Rxg6 ... Rg7 though.

Code: Select all

  time ply  score variation

 27168.4  27 +1.538 Rxe4, Qh1 Rxf4, Qxh3 Bd7, Qh2 Rf1, Qd2 Be8, Be2 Rg1, Qd8+ Kb7, Kd2 Rg2, Qa5 c3+, bxc3 Bb6, Qa6+ Kc7, cxb4 Bxg6, Kc1 Rd8, Nc3 Rg3, Nd1 Rd4, Qa4

 165098.0  28 +3.944 Rg7, Bg4 Rxg6, Qxh3 Rg7, Qh5 Rxe4, Bf3 Rxf4, Qh6 Rgf7, Bh5 Rf8, Qh7 Rh4, Qg6 b3, Be2 Rd8, Qf6 Rdd4, Nc3 Bb4, Qf3 Kb7, Nb5 Rd5, Nd6+ Bxd6, exd6 Rxd6

For what it is worth, the PV ends in this won position for Black:
[D] 2b5/1k6/2prp3/8/2p4r/1p3Q2/1P2B3/2K5 w - - 0 54