Author: Jeremiah Penery
Date: 22:20:45 02/17/00
Go up one level in this thread
On February 17, 2000 at 23:19:37, Laurence Chen wrote: >You are still missing the whole point and I still think that your comparison is >wrong. You are measuring a single processor and assuming that in a quad >processor machine it will compute the same way in a single processor machine. >Quad processors allow for Parallel Processing, which is much more efficient than >a single processing. This is not correct for searching a tree, like in chess. Efficiency goes down with more processors. This is why DB with ~500 processors was only getting about 30% total efficiency from them. This is also why you get _less than_ number_of_processors speedup over the single-processor. Take, for instance, Crafty running on Bob's machine (4x400MHz) vs. Crafty running on Lonnie's Kryotech Super-G machine (1GHz) - Bob's machine runs Crafty slower (according to Bob), even though he has nearly twice the processing power. >Just because there's an increase of 3.75 times in speed, >the way how the information is processed makes a huge difference. So the How so? I don't think I'm understanding you... >increase in plies may not be much, but the search area can be increased and sped >up consiredably. The search area is increased, but it also increases if you speed up a single processor. A single processor at Z MHz will search faster than a 4-processor machine at 4x(Z/4) MHz in almost all cases, because of parallel loss. >Also the available number of resources is much greater in a >quad processor than in the single processor. Most of the time, but it doesn't have to be. You can make a quad-processor machine with 16MB RAM if you want... In some cases, the memory-interleaving between processors (especially in dual-processor machines, it's usually better in quads) and other resource sharing between the processors can become a bottleneck.
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.