Joshua Baker-LePain wrote:
> On Tue, 27 Dec 2005 at 2:00pm, Chris Mauritz wrote
>
>> Has anyone here benchmarked 64-bit 4.2 against a dual core opteron
>> (or Athlon 64x2) vs a pair of physical single-core opterons? It's
>> that time of year again....ordering new workstations. 8-)
>
>
> It's not exactly what you asked for, but have a look at
> <http://www.duke.edu/~jlb17/dualcore.pdf>. I benchmarked LS-DYNA and
> matlab on a dual core Opteron based system running 4.1. For each
> test, I ran 1, then 2, then 4 identical jobs.
>
> The bottom line (as always) is that the "right" choice depends on what
> you intend to run. Codes that are CPU bound (like the structural sims
> in the above benchmarks) scale almost linearly on dual cores, while
> codes that are memory intensive (like the thermal sim) do see some
> reduction in efficiency due to the shared memory controller. Based on
> my results, I went with dual cores, as thermal sims are in the
> minority of what we run.
Thanks! That is extremely helpful. These machines are primarily going
to be used for encoding streaming media, raytracing animations, and
hacking up and applying filters to uncompressed video. Both tasks tend
to be mostly cpu bound on our current workstations. It's becoming
rather cheap to build a quad-core system with the new Opteron 180 dual
core parts. A couple of years ago, you couldn't even dream of doing
this stuff in a timely manner without a room full of SGI Octanes or
equivalent.
Cheers,
