Author: Vincent Diepeveen
Date: 07:07:00 01/03/02
Go up one level in this thread
On January 02, 2002 at 16:01:49, Dann Corbit wrote: >On January 02, 2002 at 15:54:41, David Rasmussen wrote: > >>On January 02, 2002 at 15:40:08, Dann Corbit wrote: >> >>>On January 02, 2002 at 15:34:55, David Rasmussen wrote: >>> >>>>I've tried running Crafty at my universitys multiprocessor machines, with poor >>>>results. Using 4 processors was kind of ok, but not great (~750 knps), but using >>>>many more processors (16, 24, etc.) just made it slower. How crummy is that. >>> >>>Did you do the compile with 96 processors enabled? It won't do that by default. >>>I am pretty sure you need to change the define for CPUs like this: >>>#define CPUS=96 >>> >>>Look at this in chess.h: >>> >>>#if !defined(CPUS) >>># define CPUS 1 >>>#endif >>> >>>I am guessing you will see (96/4) * 0.7 * 750K = ~12M NPS or so. >> >>I changed -DCPUS=4 to -DCPUS=16 in Makefile and put mt=16 in .craftyrc > >Hmmmm..... >Does Ultra Sparc III have an interlocked exchange? I suspect that the makefile >is defaulting to mutexes or something really icky like that. U3 is made for mass parallellism. It is crafty's split strategy which is the problem here.
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.