Author: Mike Byrne
Date: 16:44:15 11/20/03
Go up one level in this thread
On November 20, 2003 at 15:25:02, Ricardo Gibert wrote: >On November 20, 2003 at 14:26:00, Juergen Wolf wrote: > >>hi >> >>if an engine-version is 20% faster than its predecessor and nothing else has >>been changed how can this be translated into (expected) ELO-improvement. >> >>probably this was already answered in the past, but can't find the thread >>(probably wrong key-words) >> >>Running a tournament (no blitz) on 4 PC's (200 games against 5-6 engines) >>takes me around 2 weeks. >> >> >>thanks in advance >> >>kind regards juergen > >s = speedup factor >d = expected approximate elo gain per speedup doubling >log(s)*d/log(2) = approximate elo improvement > >Example: > >The rule of thumb is a program will gain about 50 elo with each doubling of >processing speed. How much elo is thereby gained with a 20% speedup? > >Let s = 1 + 20/100 = 1.2 >Let d = 50. > >log(s)*d/log(2) = log(1.2)*50/log(2) = 13.2 approximate elo improvement > >I have seen the estimate of d range between 40 and 100. The estimate was higher >in the olden days. 40 or 50 is a typical estimate nowadays. Very good - I agree with the 40 or 50 and the math. On slower machines, rated around 2000 - doubling may be worh more. On very fast machines , doubling may be worth less.
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.