Author: Ed Schröder
Date: 00:59:32 10/17/97
Go up one level in this thread
>Posted by Christophe Theron on October 16, 1997 at 15:33:14: >In Reply to: Re: Hiarcs 6 wins Blitz Nunn match against Fritz5 posted by Ed >Schröder on October 16, 1997 at 03:54:32: >On October 16, 1997 at 03:54:32, Ed Schröder wrote: >>Just one comment for Rebel9. >>Rebel9 is even more selective than Rebel8. >>That's very good for fast Pc's hitting new incredible ply depths. >>From my AUTO232 tests on relative slow Pc's (Pentium 90) I get the >>impression that this narrowed selective search pays even more. >>For Rebel9 I believe an overall engine improvement has been made of >>30-40 ELO. It's a difficult topic so I am careful here but I for the >>moment think: >>- Rebel9 (Pentium 90) improvement of 40-60 ELO above Rebel8. >>- Rebel9 (P200 or higher) improvement of 20-40 ELO above Rebel8. >>This conclusion is based after I removed from all played AUTO232 games >>all the doubles and also removed the effect of the book learner as far >>that is possible of course. >In the past it was generally admitted that doubling the speed of your >program gave you around 50 more elo points (some say 60 or even 70, but >this is not the point). >But now we suspect that this strength increase is not linear. That is, >doubling the speed of a 2000elo program can make it 2050elo. But >doubling the speed of a 2400elo program maybe gives only 2430 or so. >Let's suppose your new selection system makes your program twice as fast >as the old one to get to the same ply level. Then you can except 50 more >points on a P100 machine, but a little less on a P200. >Maybe it is what happens with Rebel9? No, this is not what I meant. Of course the strength increase (after doubling the speed) is not linear. I think this is true for every chess program otherwise DB would already have an ELO of > 5000 :) I will give it another shot. Rebel, Hiarcs, Mchess are generally known that they perform better on slow time controls than on blitz. So in blitz they may lose from the fast programs but they win on 40/2:00 level. What I noticed is that on a simple P90 Rebel9 wins games on ply depth because of the new faster search. For Rebel this is a surprising new development wouldn't you say so? Well for me it is. Then on faster machines the opponent program sees more and therefore is able to avoid mistakes better resulting in a better score. I currently am running an AUTO232 session (against R8,G3,G5,M6,H6) on 1 minute average. Results are also good sofar. Then I plan 0:30 average and so on just to satisfy my curiosity. - Ed Schroder - >- Christophe -
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.