Author: Robert Hyatt
Date: 09:00:05 05/12/04
Go up one level in this thread
I haven't looked at Aaron's log files yet, but one thing stands out on his times. For the last 3 positions my dual xeon does badly. The opteron did great. His box also does great. There is an explanation... Sometimes a particular iteration has SMP problems due to something "changing". IE the best move changes or is close to changing. And the parallel speedup at that point might look bad. Once you search through the changing, the next iteration might _fly_ along, and if you can search that deeply in parallel, you see better numbers. Obviously the opteron can't be a better parallel searcher than the xeon. Hardware has _zero_ influence on the speedup on those two machines, since one has no significant SMP advantage over the other, other than the opteron is 2x faster. Goes to show how wild SMP stuff can be. :) Would be interesting to take the opteron data, but get the times for the next-to-last iteration completed rather than the last, to see if they are closer to the xeon speedup numbers. Not that either set of numbers is _wrong_ as you-know-who will state... There is significant variance as expected...
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.