Stockfish 3 Official JA Windows/Linux builds available

[IWB]

Jim, read this:

http://www.talkchess.com/forum/viewtopi ... 44&t=47881

Thx anyhow
Ingo

On my Phenom II X6, the standard AMD64 exe is faster then the Intel popcnt exe.

Good point, I did not test that. If there is a POP Version I use that. If that is slower it is a mistake somewhere ...
Maybe I'll wait another day

BYe
Ingo

[pohl4711]

I just updated the Stockfish web site, and Jim's new builds are the official version.

Thx Daylen, I will use the 64bit "Intel" POP for the IPON in a few minutes!

Bye
Ingo

The auto-compile is still 3% faster...

Stefan

Ahh, good to know. Thats how much? 2 or 3 Elo?

I use the OFFICIAL one which will be used by 99% of the users.

Ingo

I use the best one (wich contains the official 3.0 source-code).

Stefan

[Modern Times]

Jim, read this:

http://www.talkchess.com/forum/viewtopi ... 44&t=47881

Thx anyhow
Ingo

On my Phenom II X6, the standard AMD64 exe is faster then the Intel popcnt exe.

Good point, I did not test that. If there is a POP Version I use that. If that is slower it is a mistake somewhere ...
Maybe I'll wait another day

BYe
Ingo

There isn't a lot in it, about 1%.

[Modern Times]

In Shredder GUI on my AMD Phenom II X6:

same final pv for all 4 versions, excellent. Nodes searched and nodes per sec:


Engine: AMD64 Stockfish 3 JA 64bit (512 MB)
(66.075.594) 1391

Engine: AMD64 popcnt Stockfish 3 JA 64bit SSE4. (512 MB)
(66.075.594) 994

Engine: Intel64 Stockfish 3 JA 64bit (512 MB)
(66.075.594) 1367

Engine: Intel64 popcnt Stockfish 3 JA 64bit SSE (512 MB)
(66.075.594) 1394


So, in the GUI the Intel popcount exe is faster on the position I used, barely. Very little in it.

Jim - really you probably only need to offer 2 versions, Intel and Intel popcount. Take the Intel naming off them, and call them generic exes. They work fine for AMD users... (assuming they also work on older AMDs, I can't test that.)

[IWB]

Yep

i5:

Intel POP Compile:
Total time (ms) : 2206
Nodes searched : 4176431
Nodes/second : 1893214

AMD NO POP:
Total time (ms) : 2305
Nodes searched : 4176431
Nodes/second : 1811900

Favors Intel, difference (time - fast = 100%) = 4.5% (~3-4 Elo)

AMD 8350:

Intel POP Compile:
Total time (ms) : 2938
Nodes searched : 4176431
Nodes/second : 1421521

AMD NO POP:
Total time (ms) : 2954
Nodes searched : 4176431
Nodes/second : 1413822


Favors Intel, difference (time - fast = 100%) = 0.5% (~ 1Elo?)

Phenom 2:

Intel POP Compile:
Total time (ms) : 3345
Nodes searched : 4176431
Nodes/second : 1248559

AMD NO POP:
Total time (ms) : 3267
Nodes searched : 4176431
Nodes/second : 1278368

Favors AMD, difference (time - fast = 100%) = 2.4% (+ <2Elo?)

I am testing on Phenoms. I seriously doubt that it is possible to measure the difference. I will continue with what I have as this is not worth to bother

Bye
Ingo

[IWB]

I use the best one (wich contains the official 3.0 source-code).

Stefan

Many twist to justify a non standard decision - but enjoy.

Ingo

[IWB]

Jim - really you probably only need to offer 2 versions, Intel and Intel popcount. Take the Intel naming off them, and call them generic exes. They work fine for AMD users... (assuming they also work on older AMDs, I can't test that.)

I am using AMD AND I fully support this - It is just much less trouble for ALL!

Thx for the work
Ingo

[Modern Times]

Agreed Ingo, the difference is tiny if anything.

[geots]

Jim, read this:

http://www.talkchess.com/forum/viewtopi ... 44&t=47881

Thx anyhow
Ingo

On my Phenom II X6, the standard AMD64 exe is faster then the Intel popcnt exe.

Good point, I did not test that. If there is a POP Version I use that. If that is slower it is a mistake somewhere ...
Maybe I'll wait another day

BYe





Ingo

Am I to assume you always use the popcnt executable, without checking each to be sure before you do?!


gts

[melajara]

Thanks, so far half of the test is completed and I confess I'm disappointed by the result, a meager 10 ELO increase over v2.3.1

I'm even hoping something is wrong in this testing