Author: Vincent Diepeveen
Date: 05:32:39 07/08/03
Go up one level in this thread
On July 08, 2003 at 08:24:54, George Sobala wrote: >On July 08, 2003 at 06:52:43, Vincent Diepeveen wrote: > >>Speedup as it is in all the ICCA journals from the 80s and onwards is the number >>of times you are faster out of n processors when compared to 1 processor of such >>a system. >> >>So if i would get the blessed efficiency of 15% out of 500 processors that is >> >>0.15 * 500 = 75 times speedup >> >>However to compare it to the speed of a K7 1Ghz it is interesting to express it >>in Ghz in this case: 37.5 Ghz >> >>So it effectively then 37.5 times faster than a K7 at 1 Ghz, assuming that K7 >>has a hashtable of 250GB. >> >>Most programmers talk always about speedup, but that's for a fixed amount of >>cpu's they always have. Like 2. >> >>However with many cpu's i cannot always test with the same amount of CPUs so >>speedup efficiency is a better form then to measure than speedup. >> >>Best regards, >>Vincent > >That assumes that "speedup efficiency" is a constant across n processors, for >any size of n. Is it? Seems very improbable. I see many post here a 'x' % speedup caused by killermoves and other ordering stuff that changes the b.f. and therefore exponentially influences the tree. What i trivially make is a graph of how the efficiency goes from 1 to n where n is at maximum 500. You won't find bigger machines with openMP and such latencies than that on the planet.
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.