Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: SSDF figures suggest 35 Elo per doubling of speed, not 50 or 80 Elo

Author: Roberto Nerici

Date: 00:33:22 11/21/03

Go up one level in this thread


On November 20, 2003 at 22:56:45, Russell Reagan wrote:
>
>Also, it might be that with faster hardware, a 2x speed increase might not
>amount to the same ELO increase as it does for slower hardware (hince why
>Thompson got different numbers from SSDF).

This is often suggested. Ernst Heinz and Robert Hyatt and several others have
researched this (e.g. "Crafty goes deep" in the ICCA/ICGA journal).

Another factor is how the measure is made. I think the Thompson experiments you
refer to used self-play. It seems likely to me that you would get large
improvements for depth/speed improvements when two copies of the same engine are
playing themselves. After all, the deeper version sees everything the first sees
and then a little bit more (it is more interesting to think why it sometimes
loses!)

The SSDF figures are comparing engines on a faster platform playing a range of
engines on a slower platform. They may be less sensitive to speed improvements.

Finally, there may be different results again if playing a human. I would guess
that since engines are more similar to each other than they are to humans, the
gain from a given speed improvement would be lower still against a range of
human players.

Roberto/.



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.