Author: Richard Pijl
Date: 02:23:02 07/05/03
Go up one level in this thread
>>> >>>This is a different version (19.4) so the time is not really comparable, >>>but the NPS is, and the xeon is faster than the opteron. >> >>As you're running 4 threads against 2 I don't think nps is a good comparison. >>I'd rather see the time-to-ply with the same crafty versions. Or nps with SMT >>off ofcourse. >> >>Richard. > >Why does number of threads matter? It is a dual xeon vs a dual opteron. > >It just so happens that the xeon supports hyperthreading while the opteron >does not. Should I penalize the xeon? > I think you misunderstood what I meant. In the end 'time to ply' is what matters. But as you said, when using different versions of crafty the machine performance cannot be compared that way. But nps cannot be used to compare performance when running 4 threads against 2 threads. Even if the 4 thread machine reaches higher nodecount, the 'time to ply' may be longer as more double work may be executed when running 4 threads. Perhaps the best comparison would be to take the same crafty version and compare time to ply. Of course the Xeon should run with SMT on as that makes time to ply shorter on that machine. Richard.
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.