Author: Vincent Diepeveen
Date: 15:11:07 07/19/00
Go up one level in this thread
On July 18, 2000 at 19:51:47, Chris Carson wrote: >I wonder how much better it would do? > >8x 700 = 5600 >32x 300 = 9600 > >Just a hypothetical. :) >I guess 50 points. Any other known Intel >configurations that might yeild better >results. Intel anounced a 1000 processor >configuration at one time (super computer). >I would love to see DJ 6 on that system. >I this is at Los Alamos NM. :) Actually as far as i know P.Conners ran at about 180 PIII processors some games at world champs from each 450Mhz. However it's all nonshared memory, so i have no idea how slow junior is at nonshared memory machines, but when i did some calculations for DIEP at how slow it would be at a cluster, thereby assuming for each position you want to use a shared hashtable, and not a local one, then i came out that 64 processors would perhaps be able to speedup my slow searching DIEP with a factor of about 6 or so, heavily depending upon communication speed which was at that cluster 1 usec, but still... ...even if that's 0.5usec which would make speed like 8 times or so at 64 processors, that's still really bad. So "what if you give me your hardware" is no good comparision. Idem for Deep Blue. Let them have non shared machines with thousands of processors, the more processors the bigger the problems. For sure it will forfeit in any faster game like 30 0. Just to start up the processors is a huge problem. I can imagine clusters with say 2 nodes are a smaller problem and real interesting, but when the number of nodes gets huge, then for chess search there is a real communication speed problem. >Best Regards, >Chris Carson
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.