Author: Dann Corbit
Date: 00:08:24 02/13/02
Go up one level in this thread
On February 13, 2002 at 02:12:25, Aaron Gordon wrote: >I'm sure it's common sense to a lot of you here that large hash size's + high >ram bandwith = more Nps. Well, that is definately true (even with smaller hash >size's). Here's a few benchmarks I did to compare how much bandwidth it takes to >increase x%. All tests were done on an overclocked Athlon XP 1900+ (1.6GHz), >Asus A7V266-E (KT266a) with 512mb PC2400 DDR & Windows XP. I used the Fritz6 GUI >which use's Fritz 5.32 for the Fritzmark. > >32mb hash >1850MHz, 1.3gb/sec ram >1171 FritzMarks | 1680Kn/s > >128mb hash >1850MHz, 1.3gb/sec ram >1098 FritzMarks | 1575Kn/s > >32mb hash >1850MHz, 2.2gb/sec ram >1212 FritzMarks | 1755Kn/s > >128mb hash >1850MHz, 2.2gb/sec ram >1191 FritzMarks | 1690Kn/s > > >Here it shows for the 32mb hash test that a 69.23% increase in ram bandwidth >increase's the NPS 4.46%. > >For the 128mb test a 69.23% increase in ram bandwith resulted in a 7.3% increase >in NPS. > >This should give you a rough idea of what to expect when upgrading to a fast DDR >(KT266a) based system vs a fast SDR (KT133a) or slow DDR (AMD 760, KT266, etc) >system. What exactly does a "fritzmark" measure? Without a definition it might as well be "salamanders per second"
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.