Author: Tom Kerrigan
Date: 14:18:54 03/28/03
Go up one level in this thread
On March 27, 2003 at 20:06:03, Aaron Gordon wrote: >from 1GHz as you said in your previous post. Now, for memory speed you may only >get 2-10% doubling memory speed with large hash tables (256mb+). With sub-32mb >hash tables you will see next to nothing, though. Interesting. Maybe I thought that memory speed didn't affect chess programs because memory speed changes very gradually. I mean, it's unlikely that you'd be comparing a machine with 133MHz memory to one with 66MHz memory... if doubling the speed gives you a 2-10% increase, then a typical increase like 100 -> 133MHz would give you 0-3%... >>It's not clear to me that the "fullness" of a hash table is a significant >>metric. It's the usefulness of the entries in the table that counts. Do you get >>better scores (or faster times to solutions) on test suites with really big hash >>tables? >I haven't done any test suite testing, just numerous games on FICS. With smaller >hash tables it seems the engine just looses it's "oomph" after the hashtable is >full. I see a LOT more draws when I use a hashtable 128mb or lower vs 256 and >more. I mainly use Deep Fritz 7 and Fritz 8, so you may see some different >results with other engines. Generally from what I've seen though when using a >fast CPU with tons of hash the engine just seems to pack more of a punch. Not to demean your observations, but this seems like the archetypical situation for selective observation... There are a lot of posters here who will say that program X is much stronger than program Y because of some games they watched, but when a long match is done, X draws or loses to Y... -Tom
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.