Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Crafty and NUMA

Author: Robert Hyatt

Date: 08:52:01 09/03/03

Go up one level in this thread


On September 03, 2003 at 02:24:00, Gian-Carlo Pascutto wrote:

>On September 02, 2003 at 22:34:49, Robert Hyatt wrote:
>
>>>Been working a year fulltime now :)
>>>
>>
>>So?  It took you over a year to get your parallel search working.  It took
>>me weeks.
>>
>>:)
>
>In all fairness, he did a full DTS implementation, including rewriting the
>program to a nonrecursive search, while you took an easy way out.

_I_ did the _original_ DTS implementation, don't forget.  It didn't take
me years.  I started to work on it in March of 1987 and I finished my Ph.D.
completely in August of 1988, which included months of testing to produce the
data.


>
>Diep's parallel performance does seem to be better than what you and I are
>getting.
>
>I wouldn't make that design decision for the common 2 or 4 way machines
>you are most likely to end up working with, but given that he's now
>working with 512, I'm sure it has payed off handsomely in the meantime.
>
>Likewise, optimizing a program for Opteron NUMA is a whole different animal
>than optimizing it for a 512 CPU NUMA Sun Altix.
>
>I agree optimizing Crafty for the Opteron won't take an entire rewrite,
>but please don't go claiming that 'NUMA-Crafty' will run well on any NUMA
>machine.

It _will_ "run on any NUMA machine."  If you want to say it won't run well
with 512 cpus, I agree.  But then again, it won't run well on 512 CPUS if
the machine is pure SMP either.  The issues are different.

However, Crafty _does_ have some stuff to make it run reasonably well on 32
cpu boxes.  I've already done that.




>
>--
>GCP



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.