Author: Robert Hyatt
Date: 08:25:33 10/15/99
Go up one level in this thread
On October 15, 1999 at 01:16:42, Kai Skibbe wrote: >On October 14, 1999 at 10:32:21, Georg v. Zimmermann wrote: > >>And what about the pipe to HT-memory ? uh oh, I don't think I like the idea of >>pondering on 1 computer. > >I don´t think that it´s a big problem. I have a dual celeron board and I am >testing Fritz6 Beta since 4 weeks. In engine-engine games it´s very interesting >to see full search infos of both engines. The nodes/seconds of the engines are >not significantly slower. I don´t know, if they are slower at all. So the memory >seems not to be a bottleneck. > > >Kai Fritz is probably a bad test. It has never been a "smart" program, and has always relied on fast search speeds. Which means it probably fits totally in cache except for the occasional hash references. Try programs that have a much bigger 'cache footprint'. Crafty. Hiarcs. Rebel. I definitely see things drop. Best way to test this would be to take a program (say crafty) and run a single test position thru it using 1 cpu. Search to a fixed depth. Then run that same test two times at once, so that each test gets run on a different cpu. Check the times. On the dual I have here it is slower. This same test is good on quads although the 4-way interleaving really does work...
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.