Author: Christophe Theron
Date: 10:31:51 04/29/01
Go up one level in this thread
On April 29, 2001 at 01:21:21, Feivel Avrum wrote:
>I have been testing Gambit with differing hash sizes and have come to
>conclusions I do not exactly understand. The NPS varies depending on the free
>memory, free resources, load on CPU etc. Yet the "pattern" when using different
>size hash tables remains fairly constant. The results on a machine with 256 Meg
>RAM are:
>
>HT Difference from baseline in NPS
> 1 +10K
> 3 +8K
> 6 +5K
> 12 +2K
> 24 0 (Base)
> 48 -4K
> 96 -7K
>192 -17K
>
>Can someone please explain the meaning and significance of these results.
>
>Feivel
This behaviour is absolutely normal.
What you need to know is that the decrease in NPS coming from bigger hash tables
is widely offset by the decrease in the size of the tree.
As a result, the NPS goes down, but the playing strength still goes up.
If you try with 24Kb (not Mb) of HT, you will see an even better NPS, but of
course the program will be MUCH weaker.
Christophe
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.