Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Ferret

Author: Scott Gasch

Date: 11:54:28 09/21/99

Go up one level in this thread


On September 21, 1999 at 13:54:44, Bruce Moreland wrote:

>When I came back from Germany I was lazy about taking the machine out of the
>box, so it sat in the box for some huge long period of time.  Then I got mail
>from Vincent asking me to run some test for him, and there the machine sits, in
>ply 20, having looked at 2.8 trillion nodes at approximately 890K nodes/second.

I am curious as to how in the world one gets a node rate that high?  I dont
doubt that it can be done but with my code on (fairly) fast hardware I only see
about 150k.  Even if I strip out the eval routine and make it just consider
material balance I only see about 200k or so.

I have implemented every speed enhancement i can think of except coding up the
eval in assembly language by hand (I do not know i386 assembly language well
enough for that).  For instance, I use movepools to avoid calls to malloc
altogether, have piece->location tables to avoid searching for pieces, a pawn
hash, material strengths and piece counts for each side are part of the
position...

Can anyone suggest other often used ways to boost search rate?

Thanks,
Scott



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.