Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Hash effects on knowledge vs. search

Author: Ernst A. Heinz

Date: 10:10:50 11/14/97

Go up one level in this thread


On November 13, 1997 at 22:35:43, Robert Hyatt wrote:
>
>I'd like to see data on this topic.  For example, in a typical opening
>or middlegame position, 30% hash hits is *very* good.  there is *lots*
>of stuff that is stored and *never* used again, due to the way the tree
>is traversed.  I depend on the hash stuff a lot, but in normal
>middlegame
>positions I don't see any huge performance issues at all.  IE I don't
>think
>the difference between the biggest table I can use (96mb) and the
>smallest I
>ever use (1mb) is more than 10-15% at best.  Perhaps in endgames this
>goes
>up.  I'm going to try to run a small set of positions with various sizes
>of hash tables to see what happens.  But I can't believe (except for
>some
>endings) that we are talking even 50% speed changes...

For fast time controls, your above statement is certainly true.

For tournament-type time controls, however, I know *many* positions
where
"DarkThought" really hits some kind of "hash wall" when doing about 200K
nps.

In these positions, time to solution greatly decreases (up to 50%) when
doubling the hash table size from 64MB to 128MB.

=Ernst=



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.