TalkChess.com

not sure which is the fastest C engine, but i know an candidate for the slowest: http://zeta-vintage.blogspot.de/2011/10/10-nps.html 10 NPS Move Generator is ready, Alphabeta PV Search is ready, QSearch is ready, Evaluation is ready, result is 10 NPS :) Search depth 3 needs with starting position ab...

I see you got that problem now that you switched to windows. Noooo, i dit not switch to Windows, I was just curious of the reports of Tony and Dann with these 2 seconds... Here the solution: Add the keys "TdrDelay" and "TdrDdiDelay" under "HKLM\System\CurrentControlSet\Control\GraphicsDrivers" and ...

Just saw that Windows Vista and Windows 7 have an GPU-timeout of 2 seconds:

http://msdn.microsoft.com/en-us/windows ... e/gg487368


Not sure how to prevent this...

--
Srdja

There are now windows compiles in bin folder

https://github.com/smatovic/Zeta

Thanks to Dann Corbit.

Note that this is still a buggy Alpha Version.

--
Srdja

There are now windows compiles in bin folder

https://github.com/smatovic/Zeta

Thanks to Dann Corbit.

Note that this is still a buggy Alpha Version.

--
Srdja

Zeta 0962 - Code Release updated version: https://github.com/smatovic/Zeta/tree/zeta_cl_0962 changes: - mods in move generation - simplified main loop - best move child selector - less register usage - no move sorting issues: - no castle or en passant moves - plays weak notes: - on my nv gts250 this...

Looking forward to see a Diep GPU Version....i would even buy one.

Now it is up to you to show your skills Vincent. Good luck.


--
Srdja

Simple Move Generation is fast, but legality check, storing in global and sorting is expensive: Simple loop over starting position, wo legality check, wo storing, wo sorting: 1.3 Mnps per SIMD Unit Simple loop over starting position, w legality check, wo storing, wo sorting: 500 Knps per SIMD Unit S...

@Vincent: Daniel is right with his MCS approach, it performs best on GPUs. If a MCS solution can play good chess is another question. YBW You really dont want to implement Master/Slave relations on a GPU, synching threads is a performance killer. 32 Bit Move Generator Con: board presentation needs m...

Daniel was right concerning global memory usage, i just have to delete this line and double the throughput...... // copy move to global global_moves[pid*256+n] = move; running 16*2*32 threads: with saving moves in global memory, wo legality check: nodes: 64000000 ,tnode: 0 ,ab-nodes: 0 ,movecount: 1...