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,