Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: WCCC and some NPS observations

Author: Bruce Moreland

Date: 19:20:28 06/23/99

Go up one level in this thread



On June 23, 1999 at 04:12:15, Jouni Uski wrote:

>I saw these kind of peak NPS:
>
>Cilkchess       8 000 000
>Junior          1 500 000
>Ferret          1 000 000
>Fritz           1 000 000
>Nimzo             500 000
>Rebel             250 000
>
>Interestingly Junior has overtaken Fritz in NPS. May be Fritz6 has more
>knowledge added.
>
>regards Jouni

People don't understand the relation between NPS and knowledge.

If you add knowledge, meaning extra cycles, to an evaluation function, the
program will probably process fewer nodes and search less deeply.

If you spend more time ordering moves, it will process fewer nodes and search
more deeply.

In both cases you search fewer nodes, but in the first case you have more chess
knowledge and in the second case you have the same knowledge but more actual
speed.

This seems extremely obvious to me but nobody ever comments on it, because
people are overly concerned with attaining some sort of perverse gratification
via number worship.

I think that relating NPS to chess strength in computers is like relating body
weight to physical strength in humans.  There is some vague correlation but it
is easy to imagine conditions where less is more.

bruce



This page took 0.01 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.