Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: hash tables processors and long term analysis

Author: Robert Hyatt

Date: 16:34:50 05/16/02

Go up one level in this thread


On May 16, 2002 at 05:06:45, Ricardo Gibert wrote:

>On May 15, 2002 at 21:09:14, Joe McCarro wrote:
>
>>Sometimes I have my computer analyze a position overnight.  I have an amd 1gig
>>with 256 ram and a p2 266 with 512 ram.
>>
>>Q: If I set the hash tables at say 192 for the p2-266 and 92 meg hash for the 1
>>gig athlon which will have better analysis in the morning.
>>
>>I'm wondering if once the hash is full tif the computer slows ddown so much the
>>difference in processor is effectively nullified.
>>thanks in advance.
>>Joe
>
>It is normal for a chess program to "fill up" the hash table at even Blitz time
>controls. If you consider the fact that a chess program on fast hardware can do
>1 million nodes per second, then it is easy see how the hash table fills up so
>fast at 16 bytes per hash entry.
>
>There is some degradation in hash table performance when the program is allowed
>to run overnight, but for different reasons.


Also, once the table is full, the search is definitely impacted, in terms
of speed.

Just try a position with a very small hash and then a very large one,
using (say) a 10 minute search.  The speed difference can _easily_ be a
factor of two or more faster for the larger hash table...



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.