Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Computer Hardware Performance

Author: Alberto Rezza

Date: 05:58:28 03/29/02

Go up one level in this thread


On March 28, 2002 at 17:34:17, Gian-Carlo Pascutto wrote:

>You do not need to play 1000 games to see that your computer is slower.
>
>You do need (approximately) such a great amount of games to positively
>know the difference in speed generates an elo difference of 3-10 points.

I really don't know how much one particular program may differ from other
programs in this respect, but I think that "3-10 point" already includes quite a
robust error margin - about 50%, in fact.

To say that you "need" 1000 games makes about as much sense as saying that you
"need" to test the 1:2:1 ratio of Mendel's law on the offspring of a trillion
parents, and you still have to add a margin of error.

Actually you have to experiment in order to test a theory; after the theory has
passed the tests, there may be theoretical reasons for you to expect an EXACT
result, apart from random variations that may be inherent in the model.

If you have reason to believe that a certain coin is a perfect coin, you expect
50% of heads ( +/- O(sqrt(N)) ) - until you test it enough to reach the
conclusion that it is NOT perfect.

The relationship between speed and playing strength is quite obvious to me, and
it has been measured enough. It may be 40 or 50 or 60 points per doubling of
speed, but - within reasonable error margins - you know what to expect. What
reason do you have to doubt it?

Alberto Rezza



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.