Hi,
This topic was discussed here a month or so ago.
I wonder if anyone out there has any experience they can
share about specific P4 and/or Opteron optimizations of
Linux x86 codes.
We've benchmarked several of our codes, in generic P3-
compiled versions, on P4 and Opteron. I'm curious how
much additional performance we might expect on P4 if
we use P4-specific optimizations, and on Opteron if
we use Opteron-specific optimizations.
Tim Prince, in an earlier posting, suggested that for
Opteron, it would likely be about 15%.
Obviously, it depends on the code (believe me, I know
all the caveats!). But just some sense of what the "typical"
or even the "maximum" speedups we might expect, or that you've
seen, would be helpful.
Also, whether anyone has any views on what compiler might
be best; I assume it's Intel for P4, but what might it be for
Opteron? We have experience with Absoft and PGI.
I might add that with one of our codes, using Athlon-specific
optimization (in Absoft f90) gave no perceptible performance
improvement over the generic P3 compilation when run on Athlon.
But that was several years ago.
Any observations or experience you might have would be useful.
FWIW, we're talking serial code here, not MPI.
Thanks,
-P.
--
Peter S. Shenkin Schrodinger, Inc.
VP, Software Development 120 W. 45th St., 32nd Floor
646 366 9555 x111 Tel New York, NY 10036
646 366 9550 FAX USERID: shenkin
http://www.schrodinger.com DOMAIN: schrodinger DOT com
Pre-arranged conf. calls: 702-759-8420 or 888-867-7084; passcode 646-366
|