Page 1 of 1

Fast Perft utility with Magic Bitboards and Hashing

Posted: Sun Jan 19, 2014 4:17 pm
by HumbleProgrammer
While rampaging around the Wild Wild Web, I ran across a comparatively new perft utility at https://github.com/ankan-ban/perft_cpu. It was written in the summer of 2013 by Ankan Banerjee, and there is a series of forum postings about it here: http://www.talkchess.com/forum/viewtopic.php?t=48387

The utility's speed is impressive: it is one of the fastest perft counters I've encountered and out-performed my current favorite (frcperft) by almost a 4:1 ratio on a 64-bit Windows system. After several hours of testing--including multiple test positions designed to trip up perft counters--the results have been 100% accurate out to a depth of 7.

The pre-compiled binaries (Windows only) don't use transposition tables, but the source code is adequately documented and it was simple to recompile with hashing enabled. There are also versions designed to use NVIDIA GPUs (CUDA) but I was not able to test these.

Cheers!
Humble Programmer
,,,^..^,,,

Re: Fast Perft utility with Magic Bitboards and Hashing

Posted: Tue Feb 23, 2016 4:41 am
by Juddster
Hi all - long-time lurker and first-time poster here.
Just giving this old thread a bump.

Perft utility

I was very impressed with Mr. Ankan-Ban's perft program.
I have written my own perft, which is a by-product of my forthcoming chess engine.

It's not as fast as his (at least not on a per-thread basis), but it is multi-threaded. I get about 380 Mnps on my computer.

Check it out and let me know what you think (sorry, 64-bit windows only):

https://github.com/jniemann66/juddperft

Re: Fast Perft utility with Magic Bitboards and Hashing

Posted: Fri Apr 22, 2016 8:30 pm
by thevinenator
Has anyone created a chess engine that utilizes GPU move generation?

If so, what kind of actual throughput (NPS) are you getting ?