Author: Uri Blass
Date: 03:57:25 04/14/02
Go up one level in this thread
On April 14, 2002 at 06:18:46, Slater Wold wrote: >On April 14, 2002 at 04:48:01, Uri Blass wrote: > >>On April 14, 2002 at 04:40:19, Slater Wold wrote: >> >>>On April 13, 2002 at 04:25:20, Bernhard Bauer wrote: >>> >>>>On April 12, 2002 at 18:33:16, Gian-Carlo Pascutto wrote: >>>> >>>>>On April 12, 2002 at 18:31:09, Gian-Carlo Pascutto wrote: >>>>> >>>>>>On April 12, 2002 at 18:07:26, bob o wrote: >>>>>> >>>>>>>2.I read somewhere that a 2 CPU machine would give an 80% increase in NPS. Does >>>>>>>this mean that there are diminishing returns with each extra CPU? >>>>>> >>>>>>Yes. >>>>>> >>>>>>What's even more worrying is that the latest public SMP Fritz >>>>>>(Deep Fritz) only had an increase of 30% with an extra CPU, >>>>>>not 80%. >>>>> >>>>>Little correction: what I was talking about wasn't pure NPS, since >>>>>that doesn't say anything in a parallel search, but the speedup. >>>>> >>>>>You can use a 1000-way system, get 1000 times the NPS and not go >>>>>a single precent faster. >>>>> >>>>>-- >>>>>GCP >>>> >>>>That's right. 8 CPU's give about 8-times the nps of a single CPU system. >>>>Bernhard >>> >>> >>>Hell no that isn't right. There is at LEAST a 30% penalty for SMP on an 8-way >>>box. >> >>30% panelty is about speed and not anout np's. >> >>Uri > >If there is a 30% penalty, um, then, you're going to have 30% less NPS. No Bob Hyatt explained in the past that in Crafty all processors work 100% of the time and the loss of speed is not because of one processor that does nothing but because of 2 processors that analyze the same positions in part of the time. 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.