Author: Robert Hyatt
Date: 07:55:25 04/11/03
Go up one level in this thread
On April 11, 2003 at 09:07:46, Vincent Diepeveen wrote: >On April 10, 2003 at 20:32:06, Tom Kerrigan wrote: > >Tom can you please publicly say whether you disagree or agree with this math. > >Bob claims next: > a) 30% increased speed in nps for SMT for crafty > b) 3.0 speedup out of 4 processors. > >What is his speedup out 4 threads on a dual Xeon then versus 2 threads non-SMT >on the same dual Xeon? > >If he shows a speedup in TIME do you agree then that bob's trick is >to not show how many nodes a second he needs for each ply? There is _always_ a "trick" according to you. So, what do you want to measure? I posted data showing that SMT on my dual 2.8 speeds things up both in raw NPS, and in terms of "time to solution". Since you like _neither_ of those measures, what do we try next? Size of the executables for non-SMP vs SMP? MD5SUM for each? You seem to have _no_ idea of what you want to prove, other than your stuff works and everybody else's doesn't. What does NPS per ply have to do with _anything_? I've never played in a chess tournament where the TD says "the time control is 1.1 billion nodes per move, average. > >Which would mean simply that his 30% speedup in nps is simply not true. > >>On April 10, 2003 at 11:11:52, Vincent Diepeveen wrote: >> >>>with SMT that is not the case. the second cpu in SMT delivers somewhere between >>>0% and 20%. >> >>It amazes me how simple your train of thought must have been to come up with >>this one. Was it something like this? >> >>* Hyperthreading gives you 2 logical processors >>* Hyperthreading speeds things up by 0-20% >> >>Therefore, your 2nd logical processor must be 0-20% as fast as the first. >> >>Brillant! >> >>-Tom
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.