Author: karen Dall Lynn
Date: 14:20:51 02/26/02
Go up one level in this thread
On February 25, 2002 at 23:07:03, Kevin Strickland wrote: >On February 25, 2002 at 22:55:01, K. Burcham wrote: > >>It is easy to see with the fritzmark >>test how the kns fall off with each increase in hash setting. >>But i was wondering if most here agreed with this formula. > >Well that is common for the nps to drop with increasing the hash tables. The >only thing you should be concerned with setting them to high and cause hard >drive swapping. > >When you increase the hash tables the nps will drop, but the depth should be >higher. NPS is not the most important thing in computer chess. Depth is. I read recently Stever Lopez' T-Notes (week Feb 24) in CB website, who seems to have a different opinion. According to Steve, too big hash tables may force the engine to perform futile search through lots of RAM blocks before getting to crucial info. So he even said that we'd better of under-deploying than over-deploying RAM space to hash tables. The received formula produces much smaller hash tables than I would be inclined to use. For most of blitz games 5/0 running on 1.5/2.0Ghz, the rash table sizes won't go beyond 32Mb. I tried these settings but could not sense differences clearly. Moreover, a question remains: if Hashtables have to be so small or discriminated in size, why by the use of the option "games - optimize strenght" they are always increased to the RAM size borders by Fritz 7 itself? Karen
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.