Author: Robert Hyatt
Date: 09:03:11 05/06/04
Go up one level in this thread
On May 06, 2004 at 11:35:02, Gian-Carlo Pascutto wrote: >On May 06, 2004 at 11:25:54, Robert Hyatt wrote: > >>Happens. I have hundreds of tests with no super-linear speedup from Crafty >>either. > >Of course, the parallel search there is much less efficient which makes the >chance of getting > 2 a lot smaller. Clearly not so for Cray Blitz. > >>As for DTS, when you think about how I did the splitting, you can >>figure out why super-linear speedups were very unlikely... > >Why not elaborate while you're at it? > >-- >GCP OK. In crafty, I split "here" or not at all. In CB, I could see the _entire_ tree. I could see how many moves at every ply had been searched. I could see if a node appeared to be an ALL node but something bad was going on below it that made this questionable. In short, I split at nodes where everything needed to be searched most of the time, rather than at nodes where there will be a fail-high but not on the first move. This latter is what causes super-linear speedup. It is also what makes my speedups in Crafty way more variable than the speedups in CB... Logical when you think about it. If you reliably split at ALL nodes, super-linear is nearly impossible to hit.
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.