Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Formula for hash table size

Author: Peter Fendrich

Date: 07:46:26 08/28/03

Go up one level in this thread


On August 28, 2003 at 09:55:57, Gian-Carlo Pascutto wrote:

>On August 28, 2003 at 06:20:17, Peter Fendrich wrote:
>
>>On August 28, 2003 at 05:13:37, Gian-Carlo Pascutto wrote:
>>
>>>On August 28, 2003 at 05:10:10, Peter Fendrich wrote:
>>>
>>>>IIRC Chessbase/Fritz published a formula for minimum required hash table size >in order to get optimal performance.
>>>
>>>The formula only applied to Fritz 5.32, and it's total junk for newer
>>>Fritzes.
>>>
>>>>Does anyone know where I can find it and/or other similar formulas for other
>>>>engines?
>>>
>>>I know that for ChessTiger and Deep Sjeng its as much as possible,
>>>as long as you dont start swapping to disk.
>>>
>>>--
>>>GCP
>>
>>That wasn't the answer I looked for!
>>I said minimum required not maximum.
>
>For most engines, there is no minimum, or it's very low like for example
>1 Megabyte.
>
>The optimal value is the maximum.
>
>Values inbetween work, but they will give suboptimal performance. The
>difference may be very small though, especially at blitz.
>
>--
>GCP

Let's assume that I allocate 256Mb for a 30 minutes blitz game and have enough
RAM to do so. Then there will be "no" conflicts or overwrites to mention. The
hash table "hit rate" will be the optimal for my program. Now let's lower the
size to 128Mb. I will get the same hash table "hit rate" with less RAM. If I
continue to lower the size my "hit rate" will at some point start to decrease.
When distributing memory between main hash table, KingPawn hash table, Nalimov
cashe and other tables I would like to know in advance a reasonable priority
scheme.
This is only a problem when the allocated RAM is low enough to make conflicts.
For instance the KP hash table size can be set to a more or less fixed value for
all situations but at some point it is more optimal to use some of it for the
main hash table instead.

I didn't think that the Chessbase formula would solve this but maybe give some
ideas.
/Peter



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.