a serious flaw in Houdini.

[Jimmy Huggins]

Paul,

My Houdini is currently looking at the positon after:

34.Be2

Would you prefer me to not post the lines, and only the eval.

It doesn't matter because the game is over with.

[mhalstern]

Houdini appears to be behaving normally on my quad:

I7-920 4.1 GHZ

I gave Houdini 1.5a 2 Gig of ram and 256 meg Gaviota Tablebase Hash for infinite analysis:


8/6kp/3q2p1/2b1p1P1/1p5P/1P2nQN1/1P2B3/1K6 b - - 0 1

Analysis by Houdini 1.5a x64:

34...Qf8 35.Qc6 Qd6 36.Qf3
= (0.00) Depth: 6/15 00:00:00 6kN
34...Qb6 35.Ne4 Nf5 36.h5 Bd4 37.h6+ Kh8 38.Bc4
= (-0.03) Depth: 6/16 00:00:00 8kN
34...Bb6 35.Bc4 Qd1+ 36.Qxd1 Nxd1 37.Kc2 Ne3+ 38.Kd3
= (-0.08) Depth: 6/18 00:00:00 9kN
34...Qd5 35.Qxd5 Nxd5 36.Ne4 Bd4 37.Kc2 Nf4
= (-0.13) Depth: 6/18 00:00:00 13kN
34...Qd5 35.Qxd5 Nxd5 36.Ne4 Bd4 37.Kc2 Nf4 38.Bc4
= (-0.12) Depth: 7/18 00:00:00 16kN
34...Qe6
= (-0.17 ++) Depth: 8/18 00:00:00 27kN
34...Qe6 35.h5 gxh5 36.Ne4 Bd4 37.Qxh5 Qg6 38.Qf3 Qc6 39.Qf6+ Qxf6 40.Nxf6
= (-0.18) Depth: 8/18 00:00:00 32kN
34...Qe6 35.h5 Qxb3 36.Qf6+ Kg8 37.Qxe5 Qc2+ 38.Ka1 Qa4+ 39.Kb1 Qc2+ 40.Ka1
= (0.00) Depth: 9/28 00:00:00 107kN
34...Qe6 35.h5 Qxb3 36.Qf6+ Kg8 37.Qxe5 Qc2+ 38.Ka1 Qa4+ 39.Kb1 Qc2+ 40.Ka1
= (0.00) Depth: 10/28 00:00:00 141kN
34...Qe6 35.h5 Qxb3 36.Qf6+ Kg8 37.Qxe5 Qc2+ 38.Ka1 Qa4+ 39.Kb1 Qc2+ 40.Ka1
= (0.00) Depth: 11/28 00:00:00 198kN
34...Qe6 35.h5 Qxb3 36.Qf6+ Kg8 37.Qxe5 Qc2+ 38.Ka1 Qa4+ 39.Kb1 Qc2+ 40.Ka1
= (0.00) Depth: 12/28 00:00:00 273kN
34...Qe6 35.h5 Qxb3 36.Qf6+ Kg8 37.Qxe5 Qc2+ 38.Ka1 Qa4+ 39.Kb1 Qc2+ 40.Ka1
= (0.00) Depth: 13/28 00:00:00 499kN
34...Qe6 35.h5
= (0.05 --) Depth: 14/30 00:00:00 692kN
34...Qd5 35.Ne4 Bd4 36.Bc4 Nxc4 37.bxc4 Qxc4 38.Qf6+ Kg8 39.Qd8+ Kg7 40.Qf6+ Kg8
= (0.00) Depth: 14/30 00:00:00 899kN
34...Qd5 35.Ne4 Bd4 36.Bc4 Nxc4 37.bxc4 Qxc4 38.Qf6+ Kg8 39.Qd8+ Kg7 40.Qf6+ Kg8
= (0.00) Depth: 15/32 00:00:00 1157kN
34...Qd5 35.Ne4 Bd4 36.Bc4 Nxc4 37.bxc4 Qxc4 38.Qf6+ Kg8 39.Qd8+ Kg7 40.Qf6+ Kg8
= (0.00) Depth: 16/38 00:00:00 2024kN
34...Qd5 35.Ne4 Bd4 36.Bc4 Nxc4 37.bxc4 Qxc4 38.Qf6+ Kg8 39.Qd8+ Kg7 40.Qf6+ Kg8
= (0.00) Depth: 17/38 00:00:00 3456kN
34...Qd5 35.Ne4 Bd4 36.Bc4 Nxc4 37.bxc4 Qxc4 38.Qf6+ Kg8 39.Qd8+ Kg7 40.Qf6+ Kg8
= (0.00) Depth: 18/40 00:00:00 5988kN
34...Qd5 35.Ne4 Bd4 36.Bc4 Nxc4 37.bxc4 Qxc4 38.Qf6+ Kg8 39.Qd8+ Kg7 40.Qf6+ Kg8
= (0.00) Depth: 19/40 00:00:00 8158kN
34...Qd5 35.Ne4 Bd4 36.Bc4 Nxc4 37.bxc4 Qxc4 38.Qf6+ Kg8 39.Qd8+ Kg7 40.Qf6+ Kg8
= (0.00) Depth: 20/40 00:00:01 12616kN
34...Qd5 35.Ne4 Bd4 36.Bc4 Nxc4 37.bxc4 Qxc4 38.Qf6+ Kg8 39.Qd8+ Kg7 40.Qf6+ Kg8
= (0.00) Depth: 21/40 00:00:01 20077kN
34...Qd5 35.Ne4
= (0.05 --) Depth: 22/59 00:00:03 43681kN
34...Qd5 35.Ne4
= (0.16 --) Depth: 22/59 00:00:07 83299kN
34...Qd5 35.Qf6+ Kg8 36.h5 Qxb3 37.Qc6 Qc2+ 38.Ka1 Kf7 39.h6 Ke7 40.Ne4 Qxe2 41.Qxc5+ Kd8 42.Qf8+ Kd7 43.Qf7+ Kc6 44.Qe6+ Kc7 45.Qe7+ Kc6 46.Qd6+ Kb5 47.Qxe5+ Kc4 48.Ka2 b3+ 49.Ka3 Kd3 50.Nc5+ Kc4 51.Qc3+ Kd5 52.Nxb3 Qa6+ 53.Kb4 Qb6+ 54.Ka4 Qc6+ 55.Kb4 Qxc3+ 56.Kxc3
= (0.22) Depth: 22/59 00:00:09 115mN
34...Qf8 35.Qc6 Qd6 36.Qa8 Qf8 37.Qb7+ Qe7 38.Qe4 Bd4 39.Qc6 Qc5 40.Qd7+ Kh8 41.Qd8+ Kg7 42.Qf6+ Kg8 43.Qe6+ Kh8 44.Bc4 Qf8 45.Ne4 Nf5 46.Nf6 Qa8 47.Bd5
= (0.07) Depth: 22/59 00:00:11 141mN
34...Qf8 35.Qc6 Qd6 36.Qb5 Qc7 37.Ne4 Be7 38.Bc4 Kf8 39.Nd2 Qd6 40.Kc1 Nxc4 41.Nxc4 Qd4 42.Nxe5 Qe3+ 43.Kb1 Qxb3 44.Qd7 Qe3 45.Qe6 Qg1+ 46.Ka2 b3+ 47.Kxb3 Qd1+ 48.Kc4 Qc1+ 49.Kd5 Qh1+ 50.Kd4 Qxh4+ 51.Kd3
= (0.09) Depth: 23/59 00:00:13 163mN
34...Qf8 35.Qc6 Qd6 36.Qb5 Qc7 37.Ne4 Be7 38.Bc4 Kf8 39.Nd2 Qd6 40.Kc1 Nxc4 41.Nxc4 Qd4 42.Nxe5 Qe3+ 43.Kb1 Qxb3 44.Qd7 Qe3 45.Qd5 Qg1+ 46.Ka2 Qa7+ 47.Kb3 Qe3+ 48.Ka4 Qa7+ 49.Kb5 Qb8+ 50.Ka6 Qd6+ 51.Qxd6 Bxd6
= (0.07) Depth: 24/59 00:00:17 214mN
34...Qf8 35.Qc6
= (0.13 --) Depth: 25/59 00:00:20 251mN
34...Qf8 35.Qc6 Qd6 36.Qa8 Qf8 37.Qa5 Qd6 38.Ne4 Bb6 39.Qb5 Qd5 40.Qxb4 Bd8 41.Nf6 Qh1+ 42.Ka2 Qa1+ 43.Kxa1 Nc2+ 44.Kb1 Nxb4 45.Ne4 h5 46.Kc1 Bb6 47.Bc4 Bd4 48.Kd2 Bxb2 49.Nc5 Kf8 50.Ne6+ Ke7 51.Ke3
= (0.17) Depth: 25/59 00:00:35 432mN
34...Qf8 35.Qc6
= (0.22 --) Depth: 26/59 00:00:44 548mN
34...Qf8 35.Qc6 Qd6 36.Qa8 Qd5 37.Qe8 Bf8 38.h5 Qxb3 39.Qxe5+ Kg8 40.Qe4 Kf7 41.Qf3+ Kg8 42.Qc6 Qf7 43.Ne4 Qf5 44.Bc4+ Kh8 45.h6 Ng4 46.Ka2 Ne5 47.Qd5 Nxc4 48.Qxc4 Qd7 49.b3 Qd8 50.Nf6 Be7 51.Qf4
= (0.22) Depth: 26/71 00:01:19 976mN
34...Qf8 35.Qc6 Qd6 36.Qa8 Qd5 37.Qe8 Bf8 38.h5 Qxb3 39.Qxe5+ Kg8 40.Qe4 Kf7 41.Qf3+ Kg8 42.Qc6 Qf7 43.Ne4 Qf5 44.Bc4+ Kh8 45.h6 Ng4 46.Ka2 Ne5 47.Qd5 Nxc4 48.Qxc4 Qd7 49.b3 Qd8 50.Nf6 Be7 51.Qf4 Qa5+ 52.Kb1
= (0.22) Depth: 27/71 00:01:25 1046mN
34...Qf8 35.Qc6 Qd6 36.Qa8 Qd5 37.Qe8 Bf8 38.h5 Qxb3 39.Qxe5+ Kg8 40.Qe4 Kf7 41.Nf1 Bg7 42.Qb7+ Kg8 43.Qc8+ Kf7 44.Qc7+ Kf8 45.Qc5+ Kf7 46.Qc1 Bd4 47.Nxe3 Qxe3 48.Qc4+ Kf8 49.Qxb4+ Bc5 50.Qb8+ Qe8 51.Qf4+ Qf7 52.Qe5 Be7 53.Qh8+ Qg8 54.Qc3 gxh5 55.Qc8+ Kg7 56.Qf5 Qf7 57.Qe5+ Kg8 58.Bd3
+/= (0.27) Depth: 28/71 00:01:54 1419mN
34...Qf8 35.Qc6 Qd6 36.Qe8 Qe7 37.Qa8 Qf8 38.Qa5 Qd6 39.h5 gxh5 40.Qb5 Qe7 41.Qc6 Qd6 42.Nxh5+ Kg6 43.Qf3 Qe7 44.Ng3 Kg7 45.Qc6 Qd6 46.Qb7+ Qe7 47.Qc8 Bd4 48.Nh5+ Kf7 49.Nf6 Kg6 50.Qg8+ Kf5 51.Nxh7 Kf4 52.Nf6 e4 53.Nd5+ Nxd5
+/= (0.29) Depth: 29/71 00:04:31 3436mN, tb=2
34...Nf5 35.Nxf5+ gxf5 36.h5 Qf8 37.Qd5 Bd6 38.Bd3 Kh8 39.h6 f4 40.Be4 Qd8 41.Qb7 Qe7 42.Qc8+ Qf8 43.Qe6 Qe7 44.Qf5 Bc7 45.Bd3 Bb6 46.g6 hxg6 47.Qxg6 Bc7 48.Kc2 f3 49.Kd2 Bd8 50.Ke3 f2 51.Kxf2 Qf6+ 52.Kg3 Qxg6+ 53.Bxg6 Bg5 54.h7 Bc1 55.Kg4 Bxb2 56.Kf5 Bc3 57.Ke6 Bd4
= (0.25) Depth: 29/71 00:07:58 6120mN, tb=18
34...Nf5 35.Nxf5+
+/= (0.31 --) Depth: 30/71 00:08:56 6884mN, tb=22
34...Nf5 35.Nxf5+ gxf5 36.h5 Qf8 37.Qd5 Bd6 38.Bd3 Kh8 39.h6 f4 40.Be4 Qd8 41.Qb7 Qc7 42.Qa8+ Qb8 43.Qc6 Qd8 44.Qd5 Qf8 45.Qe6 Qe7 46.Qf5 Bc7 47.g6 hxg6 48.Qxg6 Bd8 49.Bd3 Bc7 50.Be4 Bd8
+/= (0.31) Depth: 30/73 00:16:15 12673mN, tb=91

[NATIONAL12]

look at d30. you posted both -.031 and +0.31

[mhalstern]

For the negative value, you are refering to this I assume:
+/= (0.31 --)

I didn't mean to misinterpret any output. I thought for a negative eval, the - would have to be before the value.

[kgburcham]

look at d30. you posted both -.031 and +0.31

look at depth 6 7 and 8 for a negative score. Depth 30 is not showing a negative score.

kgburcham

[NATIONAL12]

hi,dont shoot me.i may well be wrong.

[kgburcham]

hi,dont shoot me.i may well be wrong.

this is not the same
(0.31 --)
(0.31)

as this
(+0.31)
(-0.31)

kgburcham

[Houdini]

this is easily checked.i was running at 8cpu.when second game is over i will give direct link.

in my WBCCC game against KAMESH near the end of game i was getting approx 0.95,as each depth increased Houdini would switch from positive eval to neg evaluations and back again,i thought at the time that it could be because of nature of game.

now i have just checked another WBCCC game where white is winning and same thing is happening.one minute it is +3.0 and next -3.0.this has been consistent from d25-29.switching from pos to negative.

You're misinterpreting the output of your chess GUI.

-- after a score means it's a "fail low" (the move is worse than the score). The PV will only show two plies: the main move and the reply that refutes it and makes it fail low.

++ after a score means it's a "fail high" (the move is better than the score). The PV will show a single ply: the main move that fails high.

In Marc's analysis you'll find examples of both.
Example of a Fail High [note only the main move is given]:
34...Qe6
= (-0.17 ++) Depth: 8/18 00:00:00 27kN

Example of a Fail Low [note only the main move and its refutation are given]:
34...Qf8 35.Qc6
= (0.22 --) Depth: 26/59 00:00:44 548mN

Both cases imply that the engine needs to think longer at the current search depth to find an exact score and the corresponding PV.

Robert

[Kamesh]

Dear Paul,

I have never experienced something of that sort with my houdini. I have 1/3rd of your cores (i.e.4). I still prefer Deep Rybka as my best engine for long analysis.

There could be two aspects to your problem: 1) Using 12 cores with 24 threads (Hyperthreading enabled) and 2) the number of cores you set in the Houdini Engine Parameters. I am not a reg. at talkchess though seldom i check in here, but that's a rarity. I do not have much time to browse through forums.

Feel free to send a PM at the forum we are playing and I will try to help

Regards
Kam

[supersteve3d]

Hi Paul,

Sorry to go slightly off topic. Is this an Intel or AMD hardware setup? I'm just curious how much it costs currently for a 12 core solution.

thanks,
Steve.