Anyone have any thoughts on cpu v. cpu performance on ANSYS? Probably going to run a 24-32 core, single node system for now. Have been considering Xeon E5 of various types, but not sure where to go from there.
Thoughts and opines welcome!
cycle
How big are these simulations you're running? 32 cores seems like overkill... I haven't done super complex work with ansys, but with a 4-core i7 it never took too long.
Huge. 4-10 million cells with turbulent flow and solving energy and radiation equations. Some models are two-phase. Some are compressible even. I'm currently running 5-cores of an E5-1650 on one of our smallest models, and the result is about 1 iteration every 60 seconds.
(Firstly, excuse my lack of knowledge on Ansys) Does your version of the program support GPU acceleration. That may be a path to look at rather than getting the highest end workstation/server processor.
zanginator, it's a fledgling feature. We're currently using 14.5.7, and need to wait for a lull in analyses to switch to 15.0 (the GPU semi-supported and current version), since we can't afford to have the switch-over break currently running analyses at the moment (no time to fix them!).
The GPU acceleration is a good idea though. The results I've seen so far from Tesla cards on ANSYS is really promising, provided you're running a model they currently support. I don't think they're quite done converting all their code to be gpu-compiler friendly, though.
Just a thought, but what about the Xeon Phi? I don't know anything about ANSYS, but if it would benefit from massively parallelism on GPUs, it might be able to use stuff like the Phi.