Gingko - Hiarcs, WCCC 2015

[Ralf Müller]

In the first round of the World Computer Chess Championship Gingko found an interesting exchange sac to build a fortress.
[d]
42.Rxe4! does the job. After over 40 moves later Hiarcs agreed to a draw.

It's very interesting to see that all engines act very differently with this position. There are a couple of engines, which see the sac instantly:
- BlackMamba 2.0
- Fire 3.0 (not Fire 4.0)
- Firenzina 2.3.2
- Firenzina 2.4.1
- Maverick 1.0
- PanChess 00.611
- Rodent 1.7

Some engines don't find it immediately, but within 10 seconds:
- Matacz 1.4
- Stockfish 250714MZ++
- Stockfish 010614 Human (own setting - the default version doesn't find this in 10 seconds)

Stockfish 130615 takes 41 seconds, Komodo 6, Gull 3 and Houdini 3 don't find it in a minute.

What says your engine of choice?

[zullil]

In the first round of the World Computer Chess Championship Gingko found an interesting exchange sac to build a fortress.
[d]
42.Rxe4! does the job. After over 40 moves later Hiarcs agreed to a draw.

It's very interesting to see that all engines act very differently with this position. There are a couple of engines, which see the sac instantly:
- BlackMamba 2.0
- Fire 3.0 (not Fire 4.0)
- Firenzina 2.3.2
- Firenzina 2.4.1
- Maverick 1.0
- PanChess 00.611
- Rodent 1.7

Some engines don't find it immediately, but within 10 seconds:
- Matacz 1.4
- Stockfish 250714MZ++
- Stockfish 010614 Human (own setting - the default version doesn't find this in 10 seconds)

Stockfish 130615 takes 41 seconds, Komodo 6, Gull 3 and Houdini 3 don't find it in a minute.

What says your engine of choice?

Current developmental version of Stockfish (16 threads, 16 GB hash) takes about 3 seconds at depth 19:

Code: Select all

info depth 19 seldepth 30 multipv 1 score cp -338 nodes 18130712 nps 6777836 hashfull 9 tbhits 0 time 2675 pv e1e4 f5e4 g1f1 d6d1 f1g2 b7c6 g5e3 a6a5 c4e6 d1e2 e6c4 e2c2 e3g5 c6d6 g5h6 c2c3 h6e3 c3e1 c4f7 e1b4 f7c4

I doubt that any engine is "playing for a fortress." They simply see that all other moves lose quickly. Does any engine evaluate this position at 0.00?

[Waschbaer]

Komodo 9.1: 3s

[-3.76] d=19 41.Rxe4 fxe4 42.Be3 a5 43.Kh2 Kc6 44.Bb5 Kb6 45.Bc4 Qd8 46.Be2 Qd5 47.Bc4 Qd6 48.Bb5 Qe7 49.Kg1 Qd8 50.Bc4 Qd1 51.Kh2 Qf3 52.Bb5(0:00:03) 28564kN

[Ralf Müller]

On one thread with 1 GB Hash the latest development version takes 19 seconds (depth 24).

I doubt that any engine is "playing for a fortress." They simply see that all other moves lose quickly. Does any engine evaluate this position at 0.00?

Good point, indeed no engine evaluates this with 0.00. But anyway very interesting that a couple of weaker engines are outperforming the top engines in this position. So it doesn't seem to be, that they simply see that all other moves lose more quickly.

[Ralf Müller]

Thanks! What are your system data? How many Cores and Hash?

[Waschbaer]

AMD 8 cores, 16GByte hash (my config for analyse mode)

[Ralf Müller]

I'm curious to see Komodo 9.1's result on 1 core with 2 GB Hash. Could you please test it?

[Waschbaer]

Komodo 9.1 1 core (AMD 8350). 2 GB hash

[-3.66] d=25 41.Rxe4 fxe4 42.Be3 a5 43.Bb5 Kb6 44.Bc4 Qd1 45.Kg2 Qf3 46.Kh2 Kc6 47.Kg1 Qd1 48.Kh2 Qb1 49.Kg2 Kb6 50.Kh2 Qc2 51.Kg2 Qd1 52.Kh2 Qf3 53.Kg1 Kc6 54.Kh2 Qd1(0:00:18) 30337kN
[-3.66] d=24 41.Rxe4 fxe4 42.Be3 a5 43.Bb5 Kb6 44.Bc4 Qd1 45.Kg2 Qf3 46.Kh2 Kc6 47.Kg1 Qd1 48.Kh2 Qb1 49.Kg2 Kb6 50.Kh2 Qe1 51.Kg2 Qd1 52.Kh2 Qf3 53.Bd5 Qf5 54.Kg1 Qf6(0:00:17) 28420kN
[-3.66] d=23 41.Rxe4 fxe4 42.Be3 a5 43.Bb5 Kb6 44.Bc4 Qd1 45.Kg2 Qf3 46.Kh2 Kc6 47.Kg1 Qd1 48.Kh2 Qb1 49.Kg2 Kb6 50.Kh2 Qe1 51.Kg2 Qd1 52.Kh2 Qf3 53.Bb5 Qf5 54.Bc4(0:00:16) 26866kN
[-3.66] d=22 41.Rxe4 fxe4 42.Be3 a5 43.Bb5 Kb6 44.Bc4 Qd1 45.Kg2 Qf3 46.Kh2 Kc6 47.Kg1 Qd1 48.Kg2 Kb6 49.Kh2 Qf3 50.Bb5 Qf5 51.Bc4 Qf6 52.Kg2 Qg6 53.Be2 Qd6 54.Bc4 Qf6(0:00:16) 26344kN
[-3.66] d=21 41.Rxe4 fxe4 42.Be3 a5 43.Bb5 Kb6 44.Bc4 Qd1 45.Kg2 Qf3 46.Kh2 Kc6 47.Kg1 Qd1 48.Kg2 Kb6 49.Kh2 Qf3 50.Bd5 Qf5 51.Bc4 Qf6 52.Bd5 Qg6 53.Bc4 Qf5 54.Bd5(0:00:16) 26200kN
[-3.69] d=21 41.Rxe4 fxe4 42.Be3 a5 43.Bb5 Kb6 44.Bc4 Qd1 45.Kg2 Qf3 46.Kg1 Qf6 47.Kg2 Qf8 48.Kh2 Qf5 49.Kh1 Kc6 50.Kg2 Kd6 51.Kh2 Qf8(0:00:15) 25883kN
[-3.62] d=20 41.Rxe4 fxe4 42.Be3 a5 43.Bb5 Kb6 44.Bc4 Qd1 45.Kg2 Qf3 46.Kg1 Qf6 47.Kg2 Qf8 48.Kh2 Qf5 49.Kh1 Kc6 50.Kg2 Kd6 51.Kh2 Qf8 52.Kg2(0:00:15) 25821kN
[-3.57] d=20 41.Rxe4(0:00:15) 25734kN
[-4.07] d=20 41.Be2 Bf3(0:00:14) 23185kN
[-3.74] d=20 41.Be2 Bf3(0:00:12) 19364kN
[-3.52] d=20 41.Be2 Bf3(0:00:08) 14054kN
[-3.37] d=20 41.Be2 Bf3(0:00:07) 12444kN
[-3.27] d=20 41.Be2 Bf3(0:00:07) 11140kN
[-3.20] d=20 41.Be2 Qg6 42.Bh4 Qe6 43.Bc4 Qh6 44.Re3 Qd6 45.Be2 Qd4 46.Bf6 Kb6 47.Bf1 Qd6 48.Bg5 Ba8 49.Rd3 Qc6 50.f3 gxf3 51.Be7 Qe4 52.a5 Kxa5 53.Bxc5(0:00:06) 10251kN
[-3.13] d=19 41.Be2 Qg6 42.Bh4 Qe6 43.Bc4 Qh6 44.Re3 Qd6 45.Be2 Qd4 46.Bf6 Kb6 47.Bf1 Qd6 48.Bg5 Ba8 49.Rd3 Qc6 50.f3 gxf3 51.Be7 Qe4 52.a5 Kxa5 53.Bxc5 Bb7(0:00:06) 9665kN
[-3.11] d=18 41.Be2 Qg6 42.Bh4 Qe6 43.Bc4 Qh6 44.Re3 Qd6 45.Be2 Qd4 46.Bf6 Kb6 47.Bf1 Qd6 48.Bg5 Ba8 49.Rd3 Qc6 50.f3 gxf3 51.Be7 Qe4 52.a5 Kxa5(0:00:05) 9410kN
[-3.20] d=18 41.a5 Bc6(0:00:05) 8597kN
[-3.08] d=18 41.a5(0:00:05) 8438kN
[-3.17] d=18 41.a5 Bc6(0:00:04) 7325kN
[-3.07] d=18 41.a5 Bc6(0:00:04) 6494kN

Edit:
The time varies from 10 to 27 seconds (?)

Edit:
My fault, unloading the engine after each try will show the 15 s from above

[zullil]

On one thread with 1 GB Hash the latest development version takes 19 seconds (depth 24).

I doubt that any engine is "playing for a fortress." They simply see that all other moves lose quickly. Does any engine evaluate this position at 0.00?

Good point, indeed no engine evaluates this with 0.00. But anyway very interesting that a couple of weaker engines are outperforming the top engines in this position. So it doesn't seem to be, that they simply see that all other moves lose more quickly.

Stockfish (even newer than latest developmental version) (1 thread, 1 GB hash):

Code: Select all

info depth 22 seldepth 36 multipv 1 score cp -340 nodes 22039661 nps 2284376 hashfull 126 tbhits 0 time 9648 pv e1e4 f5e4 g1g2 a6a5 g5e3 b7b6 g2h2 d6d1 h2g2 b6c6 c4e6 c6d6 e6c4 d1e1 g2h2 d6c6 h2g2 e1b1 c4b5 c6d6 b5c4 b1d1 g2h2 d1e1 h2g2

Still doubt that any engine is using fortress detection here. They're simply settling on the best option, which happens to be much better than they realize.

[Ralf Müller]

Stockfish (even newer than latest developmental version) (1 thread, 1 GB hash):

Is this version somewhere available?

Still doubt that any engine is using fortress detection here.

I doubt this, too. But anyway there seems to be some kind of knowledge or imbalance evaluation which makes other engines to perform better than Stockfish.