This might be an old question, but does the Tesla work with prepackage Windows programs? I use ABAQUS for FEA at work, and it’s time to either convince my boss to buy a cluster or to buy me a PSC. Do programs have to be hardcoded in the CUDA environment, or will ABAQUS have access to all 960 process streams?
Any application requires a CUDA specific code path to use the hardware. Nothing is automatic.
Abaqus doesn’t have CUDA support at the moment. In the structural FEA area, I know Ansys were showing off a couple of CUDA accelerated products at one of the super computer shows, however I don’t know what the release status of that is. LSTC were talking about experimenting with CUDA acceleration of their out of core solver that LS-DYNA is based on, but again I don’t believe that has ever made it to a release product.
As both an Abaqus and LS-DYNA user with a lot of CUDA capable hardware in my cluster, it would certainly be nice to have acceleration, but to the best of my knowledge, it doesn’t exist at the moment.
I guess I’ll have to start pestering our Simulia rep about it. If enough people request it (or I’m annoying enough) I figure they’ll get on it.