Page 2 of 14

Re: Seer

Posted: Sun Oct 18, 2020 8:41 pm
by Gerd Isenberg
Frank Quisinsky wrote:
Sun Oct 18, 2020 8:29 pm
Hi Werner,

right!

Thinking on it how many differents in testing such engines we produced if all the interesting users start to compile his own one or made other important changes / mistakes and so one!

No *.exe no test!
+1

Best
Frank
I find the attitude crying for binaries boring - there sources were released today. The author already stated in the engine thread, that he will make executables available for some processors. Problem with PEXT only bitboards as well as NNUE is you best need a modern Intel processor for reasonable results, otherwise there are many combinations with/without popcount, BMI2 SSE/AVX/AVX2/AVX-512, x64!

Re: Seer

Posted: Sun Oct 18, 2020 8:48 pm
by connor_mcmonigle
Gerd Isenberg wrote:
Sun Oct 18, 2020 5:15 pm
...
Again the appeal to engine authors - please create a separate thread if introducing a new engine or important updates. The new engines 2020 thread is intended for testers. It would be interesting to see whether AMD's PEXT beats the native C PEXT implementation, most use magic bitboards as fallback, specially if people demand 32-bit binaries ;-)
...
Sorry to clutter the new engines 2020 thread. I was uncertain as to whether creating a new thread or posting in the existing new engines thread was preferred. In any case, thanks for your interest in my engine.

In my testing, compiling with PEXT disabled performs better than with PEXT enabled on my AMD CPU. This does hurt performance on AMD CPUs a fair bit as a consequence of SEE relying on a PEXT implementation for table look ups, but the lion's share of the time is still allocated to computing the NN.
mclane wrote:
Sun Oct 18, 2020 7:22 pm
Where are the executables ?!
I'm working on trying to produce binaries for Windows. On Linux, Seer is trivial to compile with CMake, though I can provide linux binaries for those who are interested. Unfortunately, using mingw64 to compile for Windows is proving problematic. I can compile without issue with MSYS2, but, the same exact code, when trying to compile natively with mingw64 results in very different behavior. I've isolated the issue to the evaluation of the network and will continue to investigate. Hopefully I'll have binaries to offer soon.

Re: Seer

Posted: Sun Oct 18, 2020 8:50 pm
by Gerd Isenberg
mclane wrote:
Sun Oct 18, 2020 8:26 pm
Gerd Isenberg wrote:
Sun Oct 18, 2020 8:14 pm
mclane wrote:
Sun Oct 18, 2020 7:22 pm
Where are the executables ?!
Good chance to learn how to build one by yourself, what is your hardware? Intel processor with BMI2 and AVX2 is suited best. AMD sucks. The bin-file is 108 MiBs, a little bit larger due to 32-bit float weights ...
I don’t know. Have a Xeon 5690 6 core intel.
A Gulftown based on Westmere microarchitecture, the 32 nm shrink of Nehalem. No BMI2, no AVX2, but SSE* and popcount, better don's test Seer with this box.

Re: Seer

Posted: Sun Oct 18, 2020 8:51 pm
by connor_mcmonigle
I was unexperienced with the difficulties of compiling on Windows and, therefore, didn't realize not providing binaries was such a sticking point. In retrospect, I definitely should have provided binaries with the release and intend to do so going forwards. My apologies!

Re: Seer

Posted: Sun Oct 18, 2020 8:58 pm
by Frank Quisinsky
Hi Gerd,

that's an argument I understand.

But for official engine testing a compile by the programmers of engines is absolute necessary.

If we have a hand full people where can do that for the community (Dann or Jim with all his experience) it will be better.
After all I saw Gabor and Günther will go in the group of people, helps with compiles.

Best
Frank

I have the software on my PC installed and try it from time to time, compare the results with the work by programmers. For myself really a boring topic because time I used more with looking in playing styles of engines. So we all have special topics or thinking ... to boring to do this one or this one.

:-)

AVX2 is super for the 10th generation from Intel.
Tomorrow I will test it with different BMI2 / AVX versions of engines on one of my i9-10900 systems.

Re: Seer

Posted: Sun Oct 18, 2020 8:59 pm
by Gerd Isenberg
connor_mcmonigle wrote:
Sun Oct 18, 2020 8:48 pm
Gerd Isenberg wrote:
Sun Oct 18, 2020 5:15 pm
...
Again the appeal to engine authors - please create a separate thread if introducing a new engine or important updates. The new engines 2020 thread is intended for testers. It would be interesting to see whether AMD's PEXT beats the native C PEXT implementation, most use magic bitboards as fallback, specially if people demand 32-bit binaries ;-)
...
Sorry to clutter the new engines 2020 thread. I was uncertain as to whether creating a new thread or posting in the existing new engines thread was preferred. In any case, thanks for your interest in my engine.

In my testing, compiling with PEXT disabled performs better than with PEXT enabled on my AMD CPU. This does hurt performance on AMD CPUs a fair bit as a consequence of SEE relying on a PEXT implementation for table look ups, but the lion's share of the time is still allocated to computing the NN.
Thanks, it was not your fault as new forum member to post in the engine thread.
Amazing that AMD's PEXT is even slower than the loop implementation. Good luck and fun with Seer's further development.
mclane wrote:
Sun Oct 18, 2020 7:22 pm
Where are the executables ?!
I'm working on trying to produce binaries for Windows. On Linux, Seer is trivial to compile with CMake, though I can provide linux binaries for those who are interested. Unfortunately, using mingw64 to compile for Windows is proving problematic. I can compile without issue with MSYS2, but, the same exact code, when trying to compile natively with mingw64 results in very different behavior. I've isolated the issue to the evaluation of the network and will continue to investigate. Hopefully I'll have binaries to offer soon.

Re: Seer

Posted: Sun Oct 18, 2020 9:23 pm
by Guenther
connor_mcmonigle wrote:
Sun Oct 18, 2020 8:48 pm
Gerd Isenberg wrote:
Sun Oct 18, 2020 5:15 pm
...
Again the appeal to engine authors - please create a separate thread if introducing a new engine or important updates. The new engines 2020 thread is intended for testers. It would be interesting to see whether AMD's PEXT beats the native C PEXT implementation, most use magic bitboards as fallback, specially if people demand 32-bit binaries ;-)
...
Sorry to clutter the new engines 2020 thread. I was uncertain as to whether creating a new thread or posting in the existing new engines thread was preferred. In any case, thanks for your interest in my engine.
No need to say sorry! You did not clutter this thread, which I introduced the same way, as the one for 2019.

Gerd is wrong in the way it would be intended for testers only, it was also created, because of easier doing research
in the future, instead of crawling through 1000s of single posts in the future.
(I am also announcing my compilations of programs sometimes, which are not updated with binaries or just source only)

The CCW engine list is practically dead already, if I once should close my chronology too, other people could create
something reliable new, much easier.

OTH Gerd is right, when he asks for additional threads from authors of new programs by themselves.

Re: Seer

Posted: Sun Oct 18, 2020 10:30 pm
by Graham Banks
Werner wrote:
Sun Oct 18, 2020 8:24 pm
+1 !
my thoughts: no .exe - no test.
+2.

Re: Seer

Posted: Mon Oct 19, 2020 12:54 am
by Terje
Guenther wrote:
Sun Oct 18, 2020 9:23 pm
OTH Gerd is right, when he asks for additional threads from authors of new programs by themselves.
I think a new subforum dedicated to threads about each engine (along with the new releases thread) would be nice.

Re: Seer

Posted: Mon Oct 19, 2020 1:38 am
by AndrewGrant
Good programs are easy to build :)