Author: Robert Hyatt
Date: 07:14:18 09/05/02
Go up one level in this thread
On September 05, 2002 at 02:31:19, Slater Wold wrote: >On September 05, 2002 at 00:47:17, martin fierz wrote: > >>thanks, this is really interesting. i never realized that SMP searches could be >>so variable in some positions and very stable in others. >>actually, vincent might like this, because it's >2 speedup :-) >> >>aloha >> martin > >I have seen on my duals, more than once, an SMP program solve a problem in 3 >minutes. > >Then 2 hours later I try the same position and it solves it in 5 minutes. > >SMP is *way* to random in its branching to be what normal people call "stable". I take that variability as "part of the pain". But if you looked at the data I gave Martin, there _is_ something that is very troubling. In one of the runs, two processors were actually _slower_ than one. Granted that it is a pathological type of position, but anytime two is slower than one gets my attention in a big way. Bruce once sent me a position where four were 1/10th as fast as one. I fooled with that for a long time before understanding it and improving the parallel search to fix it.
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.