Posted Tue, 02 Oct 2018 10:24:25 GMT by Christian Helweg Senior groundwater specialist
It seems mike-21 is able to use GPUs for modelling calculations with a very big reduction in model run times (factor 100-300).

Is this also possible in FeFlow ?

Regards Christian
Posted Thu, 04 Oct 2018 06:22:29 GMT by Peter Schätzl
Hi Christian,

at least currently this is not the case. Technically, there is one important difference that makes GPU application much more difficult (and less efficient) for FEFLOW: FEFLOW uses an implicit method, while M21 is based on an explicit formulation. The latter allows to parallely solve the equations for the single nodes, leading to the high gain factor when using GPU. We should not forget, however, that this comes with a higher price in terms of runtime beforehand, as explicit methods require a much shorter time step.

Best regards,
Posted Wed, 05 Dec 2018 16:30:07 GMT by adacovsk
I spoke with the developers of SAMG from Fraunhofer SCAI about their SAMG (Multigrid solver) a couple times and like what Peter said, it's a different type of problem. They currently use OpenMP and MPI.

The company is working on a solution to the problem that can utilize many-core CPU's, which I would imagine would use something like OpenCL. I don't think they've released anything though.

You must be signed in to post in this forum.