Author: Robert Hyatt
Date: 21:16:17 03/28/01
Go up one level in this thread
On March 28, 2001 at 19:15:43, Joshua Lee wrote: >How about all of these calculations are they in relation to elo vs humans or vs >computers? SSDF can say doubling is worth 79elo points all they want but that >doen't mean jack when their rating for fritz on slower hardware is a higher >rating than what Fritz Achieved on hardware thats over 3.4x faster!!! recheck >your data if you really want to speculate try analyzing any game of the Dutch >championship untill your copy of fritz gets plays the same move deep fritz in >the dutch championship made. You will be surprised, I have done this and if they >were running a quad 500 it is more than 3x faster than a single 500!!! >Now i cant get fritz to play all the same moves but for the most part it does >now their server has who knows how much memory while i can only use 64meg out of >192 so this may have something to do with it. This also works for Junior and >others as well. Ok so you've figured out how much time your 500 takes in >relation to their server take your ssdf rating and subtract the difference and >you sit there and tell me they are right not going to happen. > > > >On a 500Mhz P3 and 500Mhz Athlon fritz managed 2457elo out of 11 games i believe > so if you want you can call this still provisional. now a 500 is faster than a >450 K6-2 ssdf has fritz at over 2600 yea right i really believe fritz on a 450 >can play like GM Nunn > >Also Junior in dortmund recieved a 2702 provisional on a machine which was >faster than if you took the SSDF's Hardware and tripled it 2 times .....we can >take your figures or we can take fact. every triple is worth 90 points Prof. >Hyatt wasn't clear if this was human elo or computer-v computer elo but just for >this we will say it is for humans 2702-180 =2522 well they are pretty close on >this one they have junior at 2589?? But there is also estimation by doubling 79 >points ... 450 ... 900...1800...3600 well over 237points which also is around >the difference between fritz at the Dutch tournament and on a 500Mhz Athlon >Check out Tony-Hedlunds page sometime while he dosen't list every match there is >quite a bit. > > the only way their figures would be worth anything is if they added a match for >each top program against an IM or GM some 40 games at Tournament time controls >and then raised or lowerd the overall results in such a way. But that is another >message that they just wouldn't listen to so then i'd be wasting my breath. > >I think it's clear we shouldn't use any figures from SSDF. SSDF figures are _perfectly_ accurate. But only to predict outcomes between the players on the SSDF list. They have _nothing_ to do with the human Elo rating system as provided by FIDE. Other than they are computed using the same formula. But to compare a 2600 SSDF rating to a 2600 FIDE rating is simply comparing apples and oranges.
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.