Author: Robert Hyatt
Date: 07:34:33 10/17/98
Go up one level in this thread
On October 17, 1998 at 07:49:52, Kai Lübke wrote: >On October 16, 1998 at 15:52:56, Ernst A. Heinz wrote: > >>On October 16, 1998 at 11:41:21, Mark Young wrote: >>> >>>** MChess played its games so far with "dynamic TB access" activated which is >>>discouraged by Marty. It's likely it would have played better without it. >> >>Isn't it only discouraged if you use the TBs from CD? >> >>Should be not worse than for any other program that probes TBs at interior >>nodes if you copy the Tbs to your harddisk. > >I had them on HD. However, the problem seems to be that MChess slows down a lot >during heavy TB access (8 Knps or less on P6-233, about 14 Knps on PII-400) >which seems to outweigh the positive information from the TB's. >Crafty is not affected this much since it's 3-4 times faster than MChess. > >>Maybe it is worse for MChess because it extends much more aggressively, thus >>going deeper in selected lines than others and hitting on the TBs more often. > >Exactly. Crafty hardly ever goes below 1/2 its usual speed, MChess often goes to >1/10. > >--- >Shep this logic seems wrong, however. IE if I search 4x faster than Mchess, I am going to search 4x the captures, and do 4x the disk I/O as well. It might be more basic than a nps different. It *might* be that the code written by SJE does a better job at caching to avoid disk I/O than what Marty is using. But in any case, a faster program ought to get hurt more when probing in the tree than a slower program, since the fast program is going to probe more, and disk speeds are constant...
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.