Prev: Questions after install Solaris 10
Next: c0t2d0s3?
From: Rick Jones on 6 Jul 2010 19:48
adirtymindisajoyforever <getridofthespam(a)yahoo.com> wrote:
> I have a mainly CPU bound application that performs X good_things/
> second on a V490,

Just how "mainly" is "mainly?"

> I would like to predict the number when executed on a T5220? Or in
> general, is there kinda formula to convert from one sparc type
> machine to another?

If there were a magic formula, there wouldn't be nearly as much call
for benchmarking :)

If your application is single-threaded, and isn't doing much "floating
point" you might compare the SPECint2006 results between the two
systems of interest. If you happen to know that one of the
sub-benchmarks there is closest to your application, you could make
the comparison with just that rather than the full metric. CAVEAT -
make certain that the SPECint2006 results were run *without*
auto-parallelization. Otherwise it isn't a measure of single-threaded
performance.

If your application is single-threaded and is doing much "floating
point" then you could use SPECfp2006, with the same caveat.

> What if the number of CPU's is not the same?

If your application can take advantage of all the CPUs in the system
(and has no MP scaling bottlenecks of its own) then you would switch
to SPECint_rate2006 or SPECfp_rate2006.

Of course, that would just be a first-order wag. Ultimately you
really do have to run your application to find-out.

rick jones
--
Process shall set you free from the need for rational thought.
these opinions are mine, all mine; HP might not want them anyway... :)
feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...
First  |  Prev  | 
Pages: 1 2 3
Prev: Questions after install Solaris 10
Next: c0t2d0s3?