[Beowulf] Athlon64 / Opteron test

Robert G. Brown rgb at phy.duke.edu
Fri May 14 08:35:57 PDT 2004


On Fri, 14 May 2004, Joe Griffin wrote:

> Hi All,
> 
> Originally this thread was about the choice of Athlon vs. Opteron.
> But the comparison between Opteron/Intel was brought up.
> 
> I wish to state that the best choice is highly dependent
> on YOUR application.  I test various CFD and FEM

As I said earlier in the same response (and have said for years, will
write in the July CW Cluster Kickstart column and will say yet again,
I'm sure:-) YMMV, the most useful "benchmark" is your own application,
and even more, your own application compiled with the actual compilers
and linked to the actual libraries you plan to use on production and run
on a system that an exact prototype of the systems you plan to use in
production and as you note you STILL can easily see sigificant
variations in performance for many applications that can shift
cost/benefit optima by simply changing the scale of the computation!  

A benchmark running out of cache in a "small" run may scale very
nonlinearly in speed compared to the same benchmark running out of main
memory in a "large" run, or a shift in parameters can cause a good
algorithm to go bad.  The ATLAS library is a wonderful example of the
tremendous benefits that can be had just optimizing algorithm for a
given architecture, and serves equally well to illustrate that running a
benchmark with an UNoptimized BLAS may well not give you anything like
an accurate idea of real comparative system performance.

So absolutely, I agree!

On the other hand, to be fair, a lot of times one's application is
bottlenecked in just one or two core loops with a relatively simple and
repetitive structure, and it isn't that uncommon to find performance
scaling nearly perfectly with system CPU clock within an architecture
family across many variations in motherboard, bus speed, cache size, and
so forth.

But YMMV!

   rgb

> engineering applications.  I have not only seen differences
> when comparing different application programs, but also
> when comparing different uses of the same program (say if a
> person changes a job from statics to dynamics).  The biggest question
> should be how YOUR application is used.
> 
> Below is a web site comparing IA32, IA64 (linux and HPUX),  Opteron
> and an IBM P655 running AIX.   The site should only be used to
> compare hardare platforms when running our software.   I am sure
> that Fluent, LSTC/Dyna, Star-CD have similar sites.  I recomend
> finding out about the software that you will be using.
> 
> MSC.Nastran Hardware comparison:
> 
> http://www.mscsoftware.com/support/prod_support/nastran/performance/v04_sngl.cfm
> 
> Regards,
> Joe Griffin
> 
> 
> Robert G. Brown wrote:

-- 
Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu






More information about the Beowulf mailing list