Stockfish 4

[Don]

Stockfish 4 has been released. You can download at http://stockfishchess.org/ or browse the sources at https://github.com/mcostalba/Stockfish

Diffs from last version is: 43 files changed, 1534 insertions(+), 1703 deletions(-)

So 169 lines of code less but should be quite stronger than version 3. For interested people here is the breakdown of lines count.

Code: Select all

-------------------------------------------------------------------------------
Language                     files          blank        comment           code
-------------------------------------------------------------------------------
C++                             20           2110           1619           5772
C/C++ Header                    21            636            570           1558
-------------------------------------------------------------------------------
SUM:                            41           2746           2189           7330
-------------------------------------------------------------------------------

This is the first release where we have took full advantage of the power of our distributed and public testing framework. At one point we reached 41 active machines connected for a total of 136 cores and a compound testing power of 190M nps (nodes per second). This is even more noteworthy considering that Gary’s 24 core monster was offline at that moment.

But is the the ‘public’ part of our development model that is the real advantage we have and without which the improvement above SF 3 would have not been possible, no matter how hardware we throw at it.

So here is the list of developers that contributed to this release:

Code: Select all

Dan Schmidt
Dariusz Orzechowski
Eelco de Groot
Gary Linscott
jhellis3
Joona Kiiski
jundery
Leonid Pechenik
Marco Costalba
Reuven Peleg
Ryan Schmitt
Ryan Takker
Tom Vijlbrief
Uri Blass

A special thank to Daylen Yang, our precious webmaster and builder of the Mac binaries, to Peter Österlund for the Android version and to http://abrok.eu/stockfish/ from which I have, shameless, took the Windows binaries.

For the first time since Stockfish was borne, Jim Ablett didn’t compile the official builds and this is a big miss for us and for me in particular. I’d like to publicly thank him again for the help and support during all these years.

Finally a note for the user. In this version there is no more automatic detection of CPU count, so at startup Stockfish always defaults to 1 thread. In case you want to use more than one CPU you have to manually set the “Threads” UCI option accordingly.

Have Fun
SF Team

Larry has done some testing of Stockfish 4 and he is extremely impressed - especially with it's performance at long time controls but even at shorter time controls it's a massive improvement!! Very well done Marco and team!

[IWB]

Hello,

OK, to finish that topic:

Here: http://abrok.eu/stockfish/ you find 3 x64 compiles:

1. x64_modern_SSE42
2. x64_modern
3. x64

The first one is with full SS42 and is only running on 8350 and modern i3/i5/i7
The second one is just POPCNT and running on the first and AMD Phenoms2 (Maybe Athlons, can't test)
The third one is a normal x64 compile.

The first and the second have identical speeds on i5, 8350 and Phenoms2 - in other words the very first one is not needed at all.

I switched at 600 games from x64 to x64_modern. The speed difference was just 3 Elo, so I doubt anyone will recognize this in the result.

Sorry for making so much trouble, I was just concerned why this is so difficult.

Thx to the W. Battig (CEGT) he gave me the last hint to understand all that.

Bye
Ingo

[Don]

The majority of open source software is designed to make money, even if not directly. Often it is through support or value added services or simply add placements on websites.

Really? In any case, it's an irrelevant point unless you want to claim that the majority of open source chess software is designed to make money.

Please don't be a troll and ruin this thread.

[melajara]

Right now I can't test the engine properly and hope that this can be fixed.

Hello Ingo,

thanks for testing SF !

As you probably know this time I had to take care of binaries because of Jim choice to rest from computer chess. So I decided for a drastic simplification of the binaries bundled in the package. Mainly, in the case of Windows we have only 2 binaries, one is our first choice and what we recommend to test SF at its best, the other one is a fallback in _all_ the other cases, so 32 bit but also 64 bit without popcount support (sse4.2 and popcount are the same thing)...

What compiler did you use for Windows, MS Visual Studio 2012?
Would it make a real difference to build the Win64 executable (Haswell CPU) with the latest Intel compiler and the C++11 codebase ?
I'm assuming that's the best option available for maximal speed, am I right?

[reflectionofpower]

I know Stockfish 4 will pass Komodo 5 but not Houdini 3. I'll guess Vas is burning the midnight oil for Rybka 5 which I am willing to bet will be very much improved and top them all.

[Dr.Wael Deeb]

Aggressiveness 120

Cowardice 110

You'll not the only one following closely the computer chess scene
Dr.D

I have tried those settings in the past and never found them to be very good. Either Stockfish wins in brutal fashion or it gets crushed just as brutally.

Is there a place where people are trying different settings and posting results? I would like to add my own..

Peter

I am not aware of such a place Peter,but we can start a special thread here in our beloved forum.....

I can dedicate one of my i7 laptops for this purpose....
Dr.D

[Eelco de Groot]

Hello,

OK, to finish that topic:

Here: http://abrok.eu/stockfish/ you find 3 x64 compiles:

1. x64_modern_SSE42
2. x64_modern
3. x64

The first one is with full SS42 and is only running on 8350 and modern i3/i5/i7
The second one is just POPCNT and running on the first and AMD Phenoms2 (Maybe Athlons, can't test)
The third one is a normal x64 compile.

The first and the second have identical speeds on i5, 8350 and Phenoms2 - in other words the very first one is not needed at all.

I switched at 600 games from x64 to x64_modern. The speed difference was just 3 Elo, so I doubt anyone will recognize this in the result.

Sorry for making so much trouble, I was just concerned why this is so difficult.

Thx to the W. Battig (CEGT) he gave me the last hint to understand all that.

Bye
Ingo

Sorry that we did not give you the advice to try x64_modern.

I had not realized the x64__modern was an option that you had not tried yet. It was also not entirely clear what modern computers really meant, my own idea was and I posted that when somebody asked that on Rybka forum, that these are versions for which GCC can implement some version of popcount, but not the hardware popcount of SSE4.2. In that case it would still be AMD hardware popcount on your machine and assembly popcount on other machines, if possible. It is maybe not really an industry term? But I'm not sure what exactly it means. I have a 32 bit Athlon which does not apply here and no 64 bit Windows or Linux yet. So this means we have gained 3% speed for the final games!

Maybe Marco will answer about future policy what will be included for instance. If we are going to include Roman's compiles it may be better to not just pick a few of them. Marco has said the Windows compiles are borrowed from Roman for this release so no Microsoft Visual Studio compiles, to answer Mel Cooper here. Marco is on holidays still so I don't know if he will answer. Thanks to Roman for his compiles, on behalf of the Stockfish team!

Uri will be pleased about the speed-up, he might win some money! Thanks for the heads up!

Eelco

[Modern Times]

Ingo,

Sorry that we did not give you the advice to try x64_modern.

I had not realized the x64__modern was an option that you had not tried yet.

Fortunately I knew that this was the one to run on my Phenom II X6, because I had been trying some of the development versions in the last few weeks . I alerted Phenom II users on page 2 of this thread.

[mcostalba]

Ahh, I did not know that this wasn't offical. Somehow I had the impression ..., thx!

It is ok. I have took the two binaries from this site, so it is ok for you to use the x64 version for your tournament. I have not bundled the third binary, nor I have intention to do it becuase the bundling is for the casual user that could get confused by an exe galore. So I'd like, for people downloading SF, just to be very easy to understand what to do: try the best one and if it doesn't work go to the fallback one.

For people like you, that I don't consider a casual user, it is of course preferable to have the complete choice so what I missed in the announcement and what I'd like to add now is the following: the Stockfish 4 release package includes pre-compiled binaries fro the most common cases, for special requirement, like obsolete hardware, please refer to the full list of available Windows and Linux binaries at http://abrok.eu/stockfish/

[melajara]

Eelco, Thanks for the short answer, so for Windows exe the compiler is not MSVC but is it gcc (over MinGW), the Intel compiler or still something else?

I would guess the compiler producing the fastest exe is (still) the Intel compiler, so it would be normal to use this one provided the team has access to it (it's not free on Windows).