Stockfish 1.8 JA available

[snino64]

Hi Marco,

The engine is much faster to find the best plays, it's fantastic.

I note that in several positions of checkmate, the engine does not send the final play. Is this correct?
For example,

6k1/8/5KB1/6N1/8/8/8/8 w - - 0 1

34 M19 1.Nf7 Kf8 2.Bh7 Ke8 3.Ne5 Kd8 4.Be4 Kc7 5.Nc4 Kd7 6.Kf7 Kd8 7.Bc2 Kc7 8.Ba4 Kd8 9.Ke6 Kc7 10.Ke7 Kc8 11.Be8 Kb7 12.Kd8 Kb8 13.Nd6 Ka7 14.Kc7 Ka6 15.Nc4 Ka7 16.Bb5 Ka8 17.Nb6+ Ka7 18.Nc8+ Ka8 19.Bc6# (5.50)
35 (M19) 1.Nf7 Kf8 2.Bh7 Ke8 3.Ne5 Kd8 4.Be4 Kc7 5.Nc4 Kd7 6.Kf7 Kd8 7.Bc2 Kc7 8.Ba4 Kd8 9.Ke6 Kc7 10.Ke7 Kc8 11.Be8 Kb7 12.Kd8 Kb8 13.Nd6 Ka7 14.Kc7 Ka6 15.Nc4 Ka7 16.Bb5 Ka8 17.Nb6+ Ka7 18.Nc8+ Ka8 (5.90)
36 (M19) 1.Nf7 Kf8 2.Bh7 Ke8 3.Ne5 Kd8 4.Be4 Kc7 5.Nc4 Kd7 6.Kf7 Kd8 7.Bc2 Kc7 8.Ba4 Kd8 9.Ke6 Kc7 10.Ke7 Kc8 11.Be8 Kb7 12.Kd8 Kb8 13.Nd6 Ka7 14.Kc7 Ka6 15.Nc4 Ka7 16.Bb5 Ka8 17.Nb6+ Ka7 18.Nc8+ Ka8 (6.39)
37 (M19) 1.Nf7 Kf8 2.Bh7 Ke8 3.Ne5 Kd8 4.Be4 Kc7 5.Nc4 Kd7 6.Kf7 Kd8 7.Bc2 Kc7 8.Ba4 Kd8 9.Ke6 Kc7 10.Ke7 Kc8 11.Be8 Kb7 12.Kd8 Kb8 13.Nd6 Ka7 14.Kc7 Ka6 15.Nc4 Ka7 16.Bb5 Ka8 17.Nb6+ Ka7 18.Nc8+ Ka8 (6.93)
38 (M19) 1.Nf7 Kf8 2.Bh7 Ke8 3.Ne5 Kd8 4.Be4 Kc7 5.Nc4 Kd7 6.Kf7 Kd8 7.Bc2 Kc7 8.Ba4 Kd8 9.Ke6 Kc7 10.Ke7 Kc8 11.Be8 Kb7 12.Kd8 Kb8 13.Nd6 Ka7 14.Kc7 Ka6 15.Nc4 Ka7 16.Bb5 Ka8 17.Nb6+ Ka7 18.Nc8+ Ka8 (7.59)

Thank you very much to Tord, Marco, Joona and Jim.

[Tord Romstad]

I am having a problem when the cores are set to 8,the engine freezes on me.Plays fine at 4 and 6 cores on my i7 920 with HT on.Had no problems with the 1.7 series regarding setting at 8 cores though it is virtual due to HT on(Can't disable to Ht off)

This could be a bug, but in any case you should not run the program with 8 threads! Regardless of whether HT is on or off, the i7 is a quad-core CPU, and the optimal number of threads for Stockfish is 4. The N/s count will perhaps be a tiny bit higher with 8 threads (at least it is on my own i7), but this is more than balanced by the bigger search trees.

HT or not, the optimal number of threads for Stockfish (and, I assume, for most or all other chess programs) is always equal to the number of physical CPU cores.

[Frank Quisinsky]

Hi,

I was a very short time a very, very small (perhaps 0,0001%) part of this great team. Really an event for myself.

Thanks for the new version.

Best
Frank

[Nimzovik]

thanks!

[yanquis1972]

oh, i'm definitely not complaining; i just didn't imagine it was possible given sf 1.7's insane depth climbing rate.

fish 1.8 is still going, btw, d33 now from the start position, which it hit at about 1:15. again, 1 cpu.

is this some kind of new paradigm & if so, why isn't it talked about (or have i missed it). it gives the impression of a sheer brute force engine, although i realize it's not -- i mean, this thing seems built to solve chess or something.

as i typed this it hit d34, after one and a half hours.

does it just have a drastically different way of counting depth, kind of a reverse rybka (esp as it pertains to nodes)? or is this output genuine, just an insanely aggressive pruning & maybe a very nice trick or two? i mean, isn't what they did w/ 1.7 & continued here with 1.8 somewhat radical? or am i completely misinterpreting the output.

[mcostalba]

oh, i'm definitely not complaining; i just didn't imagine it was possible given sf 1.7's insane depth climbing rate.

fish 1.8 is still going, btw, d33 now from the start position, which it hit at about 1:15. again, 1 cpu.

is this some kind of new paradigm & if so, why isn't it talked about (or have i missed it). it gives the impression of a sheer brute force engine, although i realize it's not -- i mean, this thing seems built to solve chess or something.

as i typed this it hit d34, after one and a half hours.

does it just have a drastically different way of counting depth, kind of a reverse rybka (esp as it pertains to nodes)? or is this output genuine, just an insanely aggressive pruning & maybe a very nice trick or two? i mean, isn't what they did w/ 1.7 & continued here with 1.8 somewhat radical? or am i completely misinterpreting the output.

The way of counting nodes is the same of 1.7.1 and no tricks.

Also, I don't expect higher speed in reaching high depths then 1.7.1....well perhaps a bit but just a bit: no significant difference in this regards with 1.7.1

[zamar]

oh, i'm definitely not complaining; i just didn't imagine it was possible given sf 1.7's insane depth climbing rate.

fish 1.8 is still going, btw, d33 now from the start position, which it hit at about 1:15. again, 1 cpu.

Depth in modern chess engines is nothing more, but iteration counter. These monsters are just pruning and reducing sidelines so aggressively that they quickly climb in high depths.

We could also build an engine which could reach depth=100 in a fraction of a second. It would play very badly though. So reaching high depths quickly doesn't necessarily mean anything.

[swami]

[Martin Thoresen]

Swami, reading up on your SF 1.7.1 test you wrote this:

Stockfish 1.7.1 replaced Critter as new #1 with 796/1000.

So basically 1.8 scored 9 lower than 1.7.1?

[zamar]

Hi Marco,

The engine is much faster to find the best plays, it's fantastic.

I note that in several positions of checkmate, the engine does not send the final play. Is this correct?

In this version (1.8) Stockfish extracts principal variation from transposition table. This can in some cases result in wrong pv.