Author: Matthew Hull
Date: 15:44:45 08/20/02
Go up one level in this thread
On August 20, 2002 at 16:29:48, Chris Carson wrote: >On August 20, 2002 at 15:26:36, Matthew Hull wrote: > >>On August 20, 2002 at 13:41:13, Uri Blass wrote: >> >>(snip) >> >>>My impression based on looking in games of the thing that the thing did tactical >>>mistakes that the commercial of today do not do so my impression is different. >>> >>>If we look at the games they lost points or half points against humans then they >>>often did mistakes that the commercial of today do not do. >> >>This may be true, but is it not also true (and perhaps to a greater degree) that >>today's program's relative short sightedness due to lower NPS (and fewer eval >>terms per eval) also means they miss more correct moves, not necessarily >>mistakes that stand out if not done, just inferior. >> >>This is a fact that most people overlook and I think what Dr. Hyatt has been >>driving at in so many words. The compensation of these advantages outweigh the >>mistakes you are describing. >> >>Is this not a large hole in your (and Vincent's) logic which I see repeated over >>and over again in these discussions about DT/DB/DB2 versus todays progs? >> >>Regards, > >A few of obeservations about NPS. > >1. You can not compare NPS from one program to another. Evals and Searching >are handled differently, on the same HW, you can get a very wide difference in >NPS between programs. NPS is valid when comparing a specific program on >different HW. You seem to imply that NPS comparisons between programs NEVER matter, is never an issue and can't be used in this debate. I think that's just divorced from reality. > >2. If you are impressed with NPS and not results, then only DB was had higher >NPS. DJ and DF are about 2 to 3M NPS on fastest single procs and about 3 to 4M >NPS on the 8way 1Ghz box for the upcomming match. > >3. NPS may be counted differently depending on the program. > >DT/DB were fast, very fast, but they needed special HW/speed to get the results, >the commercial programs get better results than DT and about the same as DB. > >Chris Any examples to back up your statement??
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.