Best Stockfish NPS scaling yet

[zullil]

Just tested Joona's nolocks (Retire global lock) patch (see http://tests.stockfishchess.org/tests/v ... 02160ebee8 )

Best NPS scaling from 8 to 16 threads I've seen yet:

Code: Select all

Dual Intel(R) Xeon(R) CPU E5-2650 v2 @ 2.60GHz
Turbo Boost and Hyper-Threading disabled
GNU/Linux 3.18.2-031802-generic x86_64

./stockfish bench 16384 16 300000 default time 
===========================
Total time (ms) : 11100082
Nodes searched  : 278067643640
Nodes/second    : 25050954

./stockfish bench 16384 8 300000 default time 
===========================
Total time (ms) : 11100004
Nodes searched  : 163772870234
Nodes/second    : 14754307

25050954/14754307 = 1.70

Here are previous test results:

Code: Select all

C11/Stockfish/src$ ./stockfish bench 16384 16 300000 default time
===========================
Total time (ms) : 11100075
Nodes searched  : 258495860664
Nodes/second    : 23287758

C11/Stockfish/src$ ./stockfish bench 16384  8 300000 default time
===========================
Total time (ms) : 11100081
Nodes searched  : 159580768428
Nodes/second    : 14376540

23287758/14376540 = 1.62

Code: Select all

./stockfish bench 16384 16 300000 default time
===========================
Total time (ms) : 11100075
Nodes searched  : 233436761771
Nodes/second    : 21030196
 

./stockfish bench 16384 8 300000 default time
===========================
Total time (ms) : 11100001
Nodes searched  : 160664514528
Nodes/second    : 14474279

21030196/14474279 = 1.45

[Dann Corbit]

Here is a windows build of the github current source:
https://www.dropbox.com/s/lfp54yivt3tft ... sh.7z?dl=0

Tragically, the current github master branch is has now migrated to the point where I cannot use it, because logging is no longer possible (I used to be able to do a quick merge with Windiff).

Move generator is no longer compatible with the old version.

So I guess that I will be stuck on the version before the SMP NPS improvement.

But maybe someone else can use it.

[zullil]

Here is a windows build of the github current source:
https://www.dropbox.com/s/lfp54yivt3tft ... sh.7z?dl=0

Tragically, the current github master branch is has now migrated to the point where I cannot use it, because logging is no longer possible (I used to be able to do a quick merge with Windiff).

Move generator is no longer compatible with the old version.

So I guess that I will be stuck on the version before the SMP NPS improvement.

But maybe someone else can use it.

You might want to relocate your post, since unless you went to Joona's repository, your binary doesn't include the patch I tested.

[Dann Corbit]

Here is a windows build of the github current source:
https://www.dropbox.com/s/lfp54yivt3tft ... sh.7z?dl=0

Tragically, the current github master branch is has now migrated to the point where I cannot use it, because logging is no longer possible (I used to be able to do a quick merge with Windiff).

Move generator is no longer compatible with the old version.

So I guess that I will be stuck on the version before the SMP NPS improvement.

But maybe someone else can use it.

You might want to relocate your post, since unless you went to Joona's repository, your binary doesn't include the patch I tested.

This version is from: https://github.com/zamar/Stockfish/tree/nolocks
https://www.dropbox.com/s/zkwwi509p29k7 ... ks.7z?dl=0

[Jouni]

But I got only 1,72 from 1 cpu to 2 cpu - no good.

[zullil]

But I got only 1,72 from 1 cpu to 2 cpu - no good.

How did you make your measurements? Time? Hash size? Method? Hardware? OS?

[Jouni]

Windows version and bench command from console. I re-tested with more runs and got now as average 1,81 for dev version and 1,91 for SF6. Latter has 16% faster nps and easily beats dev version of course.

[zullil]

Windows version and bench command from console. I re-tested with more runs and got now as average 1,81 for dev version and 1,91 for SF6. Latter has 16% faster nps and easily beats dev version of course.

Thanks, but your response is too lacking in specifics for me to fully understand.

What parameters did you use with the bench command?

What do you mean by the "dev version"? The binary in this thread supplied by Dann? Which one? Or did you get some other binary?

Just want to be sure we are comparing oranges to oranges here, as much as possible given our differing systems.

[neelbasant]

Although newer one is lower in NPS but during analysis it is better than SF6.

Neel

[zullil]

Although newer one is lower in NPS but during analysis it is better than SF6.

Neel

Thanks, but which "newer one"? And what does "during analysis it is better" mean? Is there number you can point to?