Author: blass uri
Date: 10:14:50 09/27/99
Go up one level in this thread
On September 27, 1999 at 09:15:34, Robert Hyatt wrote: >On September 27, 1999 at 05:03:54, Jouni Uski wrote: > >>Fritz's rating is 2681. I think this is too much. Of course Fritz plays at GM >>level, but not at 2681. So there are 2 possible reasons: >>1) SSDF human games are from beginning of 1990 and probably humans know >>computers now much better or >>2) comp-comp play increases rating differences. >> >>I think, that probably both reasons are inflating list now. >>But still many thanks for SSDF for new list - we haven't anything better. >> >>Jouni > >If you interpret the list correctly, the ratings are ok. We just can't take >SSDF ratings as anywhere within 200 points of the FIDE scale... > >But for comparing two programs, it is statistically sound.. I think that the surprising fact in the list is that there that the hardware gap K6-450 vs p200 does not seem to give less benefit relative to p200 vs p90 I expected that hardware is going to give less benefit because the result of some games is decided by the opening and not by the engines(I saw 2 games Junior5-Hiarcs7.32 when the evaluation of white out of book was 0.00 and the position was a clear draw(the best for white is repetition). I expected that when programs are using faster hardware there will be more cases when the result is decided by the opening books and this is the reason that I expected less advantage from fast hardware. I think that we are usually not going to see results of 40:0 between top programs even if the hardware of one program is 1000000 times faster because of this reason. The bad book of top programs and the fact that they do not use time to calculate if to play book moves is the main reason that I expected less benefit from fast hardware. Uri
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.