Author: Robert Hyatt
Date: 08:04:51 07/17/04
Go up one level in this thread
On July 17, 2004 at 05:01:09, Sune Fischer wrote: >On July 16, 2004 at 22:07:38, Robert Hyatt wrote: > >>Right. But don't forget that you are looking at NPS. That is not parallel >>search efficiency. NPS might be 4x. Actual search will be maybe 3.1X faster, >>or something pretty close to that... > >If Omid used a 32 bit compiler crafty's effective hardware advantage would have >been: >parallel_speedup * 32to64_speedup = 3.1 * 1.4 = 4.3 > >-S. That's not my problem. IE he ran on a 64 bit processor. So _hardware_ advantage is easy to compute. If he didn't compile it correctly, there's little I can do about that. But the raw _hardware_ advantage is easy to compute. Also the 1.4 is wrong, because gcc is worse than microsoft's compiler by at least 10%. There are any number of bad things someone _could_ do to make their program run worse. IE write it in Java. But that doesn't affect the _hardware_ advantage whatsoever. The issue is/was "How much faster would crafty run on the quad than it would on Omid's single-cpu machine and would the O/S influence that at all?" The answer is that the O/S doesn't directly influence the number, but the better compiler available under XP would certainly help at least 10% or more. Our quad would be maybe 3x faster than the single cpu box Falcon ran on, if that. With windows, we'd pick up more, and since the windows XP-64 system has functional NUMA support where our Linux system did not, we'd pick up yet another 10% on the quad had we run under XP.
This page took 0.01 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.