Author: Sune Fischer
Date: 20:13:33 02/08/04
Go up one level in this thread
On February 08, 2004 at 20:43:05, Thomas Mayer wrote: Hi Thomas >Hi Sune, > >> I don't like the comparison at aceshardware, they compare nps directly >> between Diep on the Athlon with one thread and the P4 with two threads. >> I think they should have mentioned that the efficiency of the extra thread >> isn't 100%. >> So I'm left asking, "What does the test show?" and "How much faster is the >> Athlon for Diep?". > >well, that's simple math: I don't have the numbers of acehardware in mind >anymore, but let's make an example: > >Think that we have 2 Threads producing 200 knps. Now we device that through the >number of threads and get 100 knps... Multiplicate that with a guessed >efficiency of - hm - let's say a conservative 1.7. Now we have 170 knps. So in >comparison 200 knps with 2 Threads would be equal to 170 knps on a single CPU. >So you can do now the same thing with the numbers at acehardware. > >For hyperthreading we can now even create a formula how much speed up in nps >would be needed to have some payback for the hyperthreading: > >(1+s)/2 * 1.7 > 1 (s is speedup in %) >1+s > 2/1.7 >s > 2/1.7-1 >s > ~ 17.6 % > >So with a efficiency of 1.7 you must get more then 17.6% speedup in nps because >of hyperthreading that you have some payback. If the efficiency is better, let's >say 1.8: > >s > 2/1.8 - 1 >s > ~ 11.1 % > >So with a efficiency of 1.8 a speedup of 11.1% would be already enough. > >I hope I have not done to much wrong mathematic stuff above -> it's long time >ago I did it... :) I think you just proved my point :) Ace's should not assume that their average reader is capable or even aware that some additional number interpretation is needed. -S. >Greets, Thomas
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.