Author: Aaron Gordon
Date: 13:51:10 12/07/05
Go up one level in this thread
On December 07, 2005 at 16:39:16, Ed Murak wrote: >At http://www.talkchess.com/forums/1/message.html?467829 you collated these >results - > >... >124651 = AMD Opteron 275, 2.2ghz, 32 bit (WIN x64) >124651 = Athlon 64 3500+ 2.2GHz 32 bit >... >110801 = Pentium-M 2.0GHz 32 bit >110801 = Pentium 4, 3.6 ghz,560 Prescott, 32 bit (WIN/XP) >110801 = Athlon 64 3200+ 2GHz 32 bit >... > >Does it not seem a little strange that the precisely, exactly same "nps" is >reported for widely-differing CPUs (Athlon and Pentium 4 and Pentium M, say)? > >In just 17 results, 3+2=5 exact collisions with such big numbers involved is >beyond a coincidence. > >Is there some granularity in the test that we don't know about? If not, is >"nps" reported always inversely proportional to "time" reported, for the same >initial conditions and 3988843 node count? If not, would "time" be a better >measure? > >I It is possible time IS a better measure, but we had to collect the results as-is to figure this out. :) It looks like it is just rounding to the nearest second. This would be ok for longer runs (maybe 200 seconds?), but my PC took just over 15 seconds. This is a bit of a difference I think, and would like to see more accurate results. Only problem would be having everyone now re-run it and calculate the NPS by hand from the time in ms and total node count.
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.