Stockfish 4

[Henk]

Reduction is like playing hide and seek and you count 1,2,3,4,5,6,7,8,9, hundred. So you are not waiting hundred seconds, but only ten.

[fern]

Indeed!!!
We are leaches of these talented people.

Fern

[Modern Times]

The http://stockfishchess.org/download/ page boasts "Precompiled binaries for all operating systems." It should be made clearer that Linux compilations can be found by following the "Dev Versions" link.

And also, there is no x64 compile in the Windows download package that works on AMD Phenom II. That should be remedied. The one that works and they should include is stockfish_13082009_x64_modern.exe

[fern]

Ok, now I understand what YOU say, but some other guys here, including one of the fathers of the program, say it is not so....
Just semantic?

Fern

[fern]

Then that is what I said: I said that stockfish surely do a very strong pruning. Some branches are cut very soon or even at once, that I know, is proper of almost every program since the old full search width became obsolete. I even know that the last programmers stuck with that technique were the Spracklen. And even they did cut a bit.
So to me the number of ply I assume indicates how deep it goes the search in the longest branch of the tree.

Fern

[Henk]

Ok, now I understand what YOU say, but some other guys here, including one of the fathers of the program, say it is not so....
Just semantic?

Fern

At least these 30 ply are not real old school ply that cannot be. They use reductions and or extensions. Some reductions are safe some are quite risky. Could be that they only use safe reductions, but only small reductions are safe.

[edit so these reductions cannot be safe]

[kgburcham]

Stockfish Bench Intel 3970x
I assume this is one core

Total time <ms> : 2001
Nodes searched : 4132371
Nodes/second : 2065154

Everything works fine for me.
Using the Shredder interface.
The test position was with 12 min split depth
It was faster with 10 min split depth.
Using 6 cores/12 threads

Here is an example,
[D] 1r3rk1/2qb1p1n/p2b1Pp1/2p1pNPN/3pP1Q1/1P1p1R2/P5B1/n1B3K1 w - -

Engine: Stockfish 4 64 SSE4.2 (4096 MB)
by Tord Romstad, Marco Costalba and Joona
28/62 1:13 +2.86++ 27.Qh4 Bxf5 28.exf5 gxh5 29.Qxh5 Rfd8 30.Rh3 Kf8 31.Qh6+ Ke8 32.Qxh7 Nc2 (911.052.119) 12391
28/62 1:24 +2.94++ 27.Rh3 c4 28.Qh4 d2 29.Bxd2 Rxb3 30.axb3 gxf5 31.Ng7 (1.063.274.256) 12528
28/62 1:27 +3.07++ 27.Rh3 c4 28.Qh4 d2 29.Bxd2 Rxb3 30.axb3 gxf5 31.Ng7 (1.105.299.181) 12619
28/62 1:31 +3.25++ 27.Rh3 c4 28.Qh4 d2 29.Bxd2 Rxb3 30.axb3 gxf5 31.Ng7 (1.166.817.799) 12735
28/62 1:35 +3.52++ 27.Rh3 c4 28.Qh4 d2 29.Bxd2 Rxb3 30.axb3 gxf5 31.Ng7 (1.222.791.551) 12842
28/62 1:39 +3.93++ 27.Rh3 c4 28.Qh4 d2 29.Bxd2 Rxb3 30.axb3 gxf5 31.Ng7 (1.285.865.444) 12962
28/62 1:44 +4.54++ 27.Rh3 c4 28.Qh4 d2 29.Bxd2 Rxb3 30.axb3 gxf5 31.Ng7 (1.367.080.316) 13090
28/62 1:51 +5.45++ 27.Rh3 c4 28.Qh4 d2 29.Bxd2 Rxb3 30.axb3 gxf5 31.Ng7 (1.477.070.515) 13233
28/62 2:03 +6.82++ 27.Rh3 c4 28.Qh4 d2 29.Bxd2 Rxb3 30.axb3 gxf5 31.Ng7 (1.653.645.027) 13410
28/62 2:16 +8.87++ 27.Rh3 c4 28.Qh4 d2 29.Bxd2 Rxb3 30.axb3 gxf5 31.Ng7 (1.860.818.847) 13604
28/62 2:35 +11.95++ 27.Rh3 c4 28.Qh4 d2 29.Bxd2 Rxb3 30.axb3 gxf5 31.Ng7 (2.153.935.921) 13807
28/62 3:04 +15.65 27.Rh3 Bxf5 28.exf5 e4 29.Qh4 Bh2+ 30.Rxh2 d2 31.Bxd2 Qxh2+ (2.579.240.151) 14006
29/62 3:15 +15.73++ 27.Rh3 Bxf5 28.exf5 e4 29.Qh4 Bh2+ 30.Rxh2 d2 31.Bxd2 Qxh2+ 32.Kxh2 Rfe8 (2.747.838.171) 14040
29/62 3:20 +15.81++ 27.Rh3 Bxf5 28.exf5 e4 29.Qh4 Bh2+ 30.Rxh2 d2 (2.824.375.555) 14059
29/62 3:25 +15.93++ 27.Rh3 Bxf5 28.exf5 e4 29.Qh4 Bh2+ 30.Rxh2 d2 31.Bxd2 Rxb3 (2.889.536.128) 14084
29/62 3:32 +16.12++ 27.Rh3 Bxf5 28.exf5 e4 29.Qh4 Bh2+ 30.Rxh2 d2 31.Bxd2 Qxh2+ (2.992.822.210) 14103
29/62 3:39 +16.39++ 27.Rh3 Bxf5 (3.093.404.855) 14112

[Uri Blass]

Sorry, still above my head. What that reduction you mention is?
In any case my common sense say to me that if old dedicate units reached 5 or 6 ply in middle game, cannot be that modern engines running in hardware dozens of thousands of times faster do the very same thing.

In any case, it is a technical matter that that not touch me very deeply. I am beaten these days by 0 ply engines....

Like Louis said. Henk is a bit confused. The 30 plies you mention are really 30 plies : Stockfish searches one ply, then, with iterative deepening, goes for a 2-plies search, then a 3-plies search, and so on, until it has no time anymore. And yes, it's a great achievement to reach such a depth in so little time. It means that Stockfish has a great move ordering and does a lot of pruning. In a clever way.

Stockfish is a great achievement but
I disagree that it is a great achievement to reach a big depth in a short time.

The target is not to get bigger reoported depth and
it is clearly possible to get bigger depth than stockfish by doing more aggresive pruning and be clearly weaker.

[fern]

I suppose that it is in this side of the equation where smart programming comes to the scene. The ponts is To know what to cut down at once or soon without cutting down the critical line. That suppose smart search technique and smart knowledge about where and when to say "stop the search of this line because it is irrelevant"

Fern

[JuLieN]

Stockfish is a great achievement but
I disagree that it is a great achievement to reach a big depth in a short time.

The target is not to get bigger reoported depth and
it is clearly possible to get bigger depth than stockfish by doing more aggresive pruning and be clearly weaker.

Exactly. That's why I said "clever". I already made versions of my engine that reached enormous depths and... played even worse than the released versions already do (yes it's possible!)

The pruning in Stockfish is done because tests show it increases its strength. Tuning this is very tricky. And it relies on a very good move ordering. If your move ordering was perfect, you'd always only had to play the first move each time, discarding all the other ones. Something that never happens. But still, getting the first move right occurs quite often anyway. (Note that if move ordering was really perfect, you wouldn't even need to search the tree, as you'd get it right from the root. )