Komodo shows TB hits in starting position

[Jouni]

Analysis by Komodo 9.2 64-bit:

1.d4 Nf6 2.c4 e6 3.Nf3 d5 4.Nc3 c5 5.Bg5 dxc4 6.e3 Nc6 7.dxc5 Qxd1+ 8.Rxd1 Bxc5 9.Bxc4 a6 10.Rc1 h6 11.Bf4 0-0 12.Na4 Ba7 13.Be2 Re8 14.Ne5 Nxe5 15.Bxe5 Bd7 16.Nc3 Rac8 17.Bf3 Bc6 18.Bxc6 Rxc6 19.Bxf6 gxf6 20.Rd1
= (0.24) Depth: 34 00:29:40 9038mN, tb=4

Quite surprising, because I have only basic 4 core PC and very few installed 6 piece tablebases! Question: what's the minimum number of plies to reach 6 piece position from start? BTW Stockfish with 53 selective plies is not probing any tablebases.

[Robert Pope]

I haven't played out an actual sequence, but I think it is only 14 ply:
4 ply to get the two queens out
10 ply to capture 5 pieces each on the queenside

[bob]

I haven't played out an actual sequence, but I think it is only 14 ply:
4 ply to get the two queens out
10 ply to capture 5 pieces each on the queenside

way off. there are 32 pieces on the board. If you could make every move a capture, you need 26 moves to reach a 6 piece ending.

[hgm]

Add 16 ply for capturing the pawns. And you would need some check evasions too.

[voyagerOne]

My guess is that K probes in Q-search.

It may have an internal EGTB, with the most common piece constellation, in order to keep memory foot-print low.

[Robert Pope]

I haven't played out an actual sequence, but I think it is only 14 ply:
4 ply to get the two queens out
10 ply to capture 5 pieces each on the queenside

way off. there are 32 pieces on the board. If you could make every move a capture, you need 26 moves to reach a 6 piece ending.

D'oh!

[mjlef]

My guess is that K probes in Q-search.

It may have an internal EGTB, with the most common piece constellation, in order to keep memory foot-print low.

Actually, Komodo does not probe TB in qsearch. The default will make it probe in the main search only. But there are lots of extensions possible in very limited PVs so that is how it can hit TB probes from the opening position (with enough processing power and time, of course).

[zullil]

BTW Stockfish with 53 selective plies is not probing any tablebases.

Latest Stockfish with 20 threads:

Code: Select all

info depth 41 seldepth 51 multipv 1 score cp 16 nodes 80301788941 nps 30393754 hashfull 999 tbhits 3 time 2642049 pv e2e4 e7e6 d2d4 d7d5 b1d2 c7c5 g1f3 g8f6 e4d5 e6d5 f1b5 c8d7 b5d7 b8d7 e1g1 f8e7 d4c5 d7c5 d2b3 c5e4 f3d4 e8g8 d4f5 f8e8 f5e7 e8e7 c2c3 d8c7 b3d4 a8e8 c1e3 c7d7 h2h3 h7h6 d1c2 e4g5 e3g5 h6g5 f2f3 d7c7 a2a4 c7e5 f1d1 a7a6 c2f2 f6d7 a4a5 d7f6 b2b4

[Eelco de Groot]

My guess is that K probes in Q-search.

It may have an internal EGTB, with the most common piece constellation, in order to keep memory foot-print low.

Actually, Komodo does not probe TB in qsearch. The default will make it probe in the main search only. But there are lots of extensions possible in very limited PVs so that is how it can hit TB probes from the opening position (with enough processing power and time, of course).

This is very interesting. I do remember from a casual observation of the TCEC games, it was just as if Komodo was reaching deeper in the endgames, where Stockfish used to have always the highest depthnumbers, in endgames this seemed to be no longer the case!

Stockfish gets its depths from the enormously aggressive Late Move Pruning, but the pruning and reductions are across the whole tree, whereas extensions will be able to always get even higher depths than reductions, but with the risk of never stopping... For extensions that go exclusively to one playing side only, the risk is limited that the search will explode. Stockfish only has left check extensions and the singular extensions. These usually are one side only extensions, in my limited opinion anyway? We already know from Rebel times and Ed, that any attempt he did to limit check extensions has failed. Even Stockfish limits them very little only, for negative SEE cases.

But with extensions that are safe, indeed, it should be possible to reach very high depths with them even when the nominal search depth/iteration number, is not so high. Interesting that Komodo has found a way, to search even deeper than Stockfish

[Eelco de Groot]

BTW Stockfish with 53 selective plies is not probing any tablebases.

Latest Stockfish with 20 threads:

Code: Select all

info depth 41 seldepth 51 multipv 1 score cp 16 nodes 80301788941 nps 30393754 hashfull 999 tbhits 3 time 2642049 pv e2e4 e7e6 d2d4 d7d5 b1d2 c7c5 g1f3 g8f6 e4d5 e6d5 f1b5 c8d7 b5d7 b8d7 e1g1 f8e7 d4c5 d7c5 d2b3 c5e4 f3d4 e8g8 d4f5 f8e8 f5e7 e8e7 c2c3 d8c7 b3d4 a8e8 c1e3 c7d7 h2h3 h7h6 d1c2 e4g5 e3g5 h6g5 f2f3 d7c7 a2a4 c7e5 f1d1 a7a6 c2f2 f6d7 a4a5 d7f6 b2b4

I was interested in the openings played. With standard Stockfish, I also got a French opening. And I believe I have seen that more often. This is just at a shallow depth, with latest master and my new computer :

Engine: Stockfish 280316 64 BMI2 MP (512 MB)
by T. Romstad, M. Costalba, J. Kiiski, G. Linscott

33/46 11:34 +0.19 1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6
5.exd5 exd5 6.Bb5+ Bd7 7.Bxd7+ Nbxd7
8.O-O Be7 9.dxc5 Nxc5 10.Nb3 Nce4
11.Nfd4 O-O 12.c3 Re8 13.Nf5 Nd6
14.Nxe7+ (3.355.303.210) 4834

best move: e2-e4 time: 12:26.844 min n/s: 4.834.577 nodes: 3.673.635.038


I also managed to compile a new version of the Rainbow Serpent, for the first time a 64 bit version, at last The new computer is still a Quad, i7so no 20 threads possible, I could try hyperthreading though. And I have more Ram available than before.

Rainbow Serpent plays a Queen's Indian, Petrosian variation, according to the Shredder opening book. 4.a3 {!?}
[D]rnbqkb1r/p1pp1ppp/1p2pn2/8/2PP4/P4N2/1P2PPPP/RNBQKB1R b KQkq -
Not something you would expect from Stockfish, or from the computer in general? I am unfamiliar with this opening but the Shredder book has lots of games with it, so once it seems was played fairly often. Nice to have it play just like the great Petrosian!

[D]rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq -

Engine: Rainbow Serpent 20162803_001 MP (i7 6400, four threads, 512 MB, latest Stockfish master, plus the usual changes)
by T. Romstad, M. Costalba, J. Kiiski, G. Linscott

17/33 0:01 +0.24 1.e4 e5 2.Nf3 Nf6 3.Nxe5 d6 4.Nf3 Nxe4
5.Qe2 Qe7 6.d3 Nf6 7.Nc3 h6 8.Be3 Nc6
9.O-O-O Be6 10.Kb1 d5 11.d4 Bf5
12.Re1 O-O-O (9.324.586) 4843
.
.
.
28/50 4:16 +0.12 1.d4 e6 2.c4 Nf6 3.Nf3 b6 4.a3 Bb7
5.Nc3 d5 6.Bf4 dxc4 7.e3 Nd5 8.Bxc4 Nxc3
9.bxc3 Bd6 10.O-O O-O 11.Ne5 Nd7
12.Qg4 Nf6 13.Qh3 h6 14.Rfc1 (1.145.889.608) 4466

29/50 4:46 +0.15++ 1.d4 (1.280.389.570) 4466

29/53 5:18 +0.12-- 1.d4 Nf6 (1.421.020.426) 4468

29/53 6:31 +0.08 1.d4 Nf6 2.c4 e6 3.Nf3 b6 4.a3 Bb7
5.Nc3 d5 6.Bf4 Bd6 7.Bxd6 Qxd6
8.cxd5 exd5 9.Qa4+ Nbd7 10.Nb5 Qc6
11.e3 a6 12.Nc3 Qxa4 13.Nxa4 c5
14.Nc3 (1.757.295.818) 4483

30/54 6:55 +0.16 1.d4 e6 2.c4 Nf6 3.Nf3 b6 4.a3 Bb7
5.Nc3 d5 6.Bf4 Bd6 7.Bxd6 Qxd6
8.cxd5 exd5 9.e3 O-O 10.Be2 c5
11.O-O Nbd7 12.dxc5 bxc5 13.b4 Rfc8
14.Nb5 (1.858.636.655) 4474

best move: d2-d4 time: 6:55.531 min n/s: 4.474.136 nodes: 1.858.636.655