Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Beowulf Crafty

Author: José de Jesús García Ruvalcaba

Date: 16:49:51 05/17/00

Go up one level in this thread


On May 17, 2000 at 18:03:46, Robert Hyatt wrote:

>On May 17, 2000 at 15:33:13, Peter Hegger wrote:
>
>>Hello
>>Bob, I was wondering how many nps you will expect to hit when you get crafty up
>>and running on the Beowulf cluster.
>>Have you an approximate time frame when you think you may have the beast going?
>>Or, is that something that is unpredictable at this point?
>>Thanks,
>>Peter
>
>
>It is unpredictable.  There are lots of things to do... but obviously it
>ought to hit 10M nodes per second.  Whether that is 10x faster than my
>normal search on a single quad is very doubtful.  But I don't think that
>hoping for 50% efficiency is out of line... which means an effective speed
>of 5M.  And this is based on the current cluster.  Another 8 quad xeons are
>due in this Summer, for a total of 18.  That just might really hit 10M useful
>nodes per second.
>
>But lots of things are left to play with, like shared vs local transposition
>tables, carefully scheduling positions on nodes that have seen them before,
>etc...

How portable will the message-passing code in crafty be?
Any hope of seeing it in action in Sun clusters, for example?
José.



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.