Author: Robert Hyatt
Date: 14:01:34 05/06/04
Go up one level in this thread
On May 06, 2004 at 16:12:33, Gerd Isenberg wrote: >On May 06, 2004 at 12:19:13, Gian-Carlo Pascutto wrote: > >>I hope you don't mean the ones blow. >> >>Are you still claiming you 'measured' 3.1 which supposedly contradicts >>the 2.8 I measured? >> > >Gian-Carlo, > >sorry, but what is your point with such chaotical, none deterministic systems? >If i imagine the chaotical effects even with one processor after some minor >code/data changes and different processor architectures. If i imagine how >chaotical multiprocessor systems are, where memory and cache coherency are big >issues. > >Did you exactly the same test with crafty on the same system as Bob? >Or did you mention your own experience with Deep Sjeng and claim in general with >multprosser systems smp or numa, speedup with nullmove is less than Bob's >suggested formula for 1 <= N <= 8: > >Nullmove disabled: 1 + (N-1)*0.7 >Nullmove enabled: 1 + (N-1)*0.6 > >Thanks, >Gerd I find this "search for the one true SMP speedup number" for Crafty to be completely ridiculous. The idea that there _is_ such a single number is silly. Much less all the arguing about "I get 2.8 you got 3.0 or 3.1 so your numbers are bad." I've just posted 9 log files from the Opteron. I have no idea how all the numbers will look as I have not gone over them yet. I ran them as I had time on a good machine and I have plans to write up the Crafty parallel search algorithm before long and that will need some reliable data. But just because my linear approximation to a probably non-linear function varies based on some specific test set really doesn't mean the formula is right, wrong or anything else... But it does seem hopeless to explain it over and over as it is just like that damned energizer rabbit, it keeps going and going and going...
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.