Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Multi-threading issues

Author: Robert Hyatt

Date: 20:03:07 11/09/02

Go up one level in this thread


On November 09, 2002 at 04:57:22, Gian-Carlo Pascutto wrote:

>On November 09, 2002 at 00:15:28, Robert Hyatt wrote:
>
>>That is not hard to implement, and it will produce a good speedup for 2-4
>>processors.
>
>It never got over 1.3 for me.
>
>--
>GCP


I don't have my dissertation handy, but PVS got better than 1.3 for everyone
that ever tried it and published results in the JICCA.  IE if I just split at
the root only, the first version of Cray Blitz got > 1.3 on average using two
cpus in 1983 at the WCCC.  In fact, 1.5 was a good average.  Unfortunately,
four processors barely improved that so that I took Monty Newborn's ande Murray
Campbell's suggestion and implemented the PVS algorithm.  If you do it right,
there is no reason it won't produce reasonable speedups for 2, but with all
processors locked together at any specific node, going beyond 2 hurts,
especially in the endgame.

I'll locate and report my original 2-cpu and 4-cpu PVS numbers when I get to
the office, they are in my dissertation.



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.