[Shootout-list] Yes! Re: Dual CPU machine?
John Skaller
skaller@users.sourceforge.net
Fri, 20 May 2005 12:26:50 +1000
On Fri, 2005-05-20 at 00:14 +0200, Shae Matijs Erisson wrote:
> Greg Buchholz <sleepingsquirrel@yahoo.com> writes:
> My desktop is a two cpu 1544MHz Athlon MP, so I'm very interested.
So am I .. do you have ANY simplish C programs that benefit
significantly?
> With the sudden upsurge of multi-core CPUs, I think lots of people will soon be
> interested in this. Can multiple CPUs can be simulated with User Mode Linux or
> something similar?
Semantically perhaps, but there is no way to 'simulate' a performance
measurement based on a real time clock, without making the kind
of assumptions which the benchmarks would actually be checking.
Ocaml, for example, can spawn threads, but they cannot
be concurrent (there is a global lock to protect the garbage
collector which isn't thread-safe).
OTOH many problems can easily be divided between processes
and use message passing -- numerical problems are typical,
where there is a very expensive calculation on a small
amount of data.
> For me, the most interesting benchmark would be "scaling".
> It would test to see how well a benchmark implementation scales from one cpu up
> to thirty two cpus.
Indeed .. but you'll need a 32 CPU machine to test do the
measurements .. the year after next we'll surely have
4 core CPUs .. :)
--
John Skaller, skaller at users.sf.net
PO Box 401 Glebe, NSW 2037, Australia Ph:61-2-96600850
Download Felix here: http://felix.sf.net