On Mon, Apr 15, 2013 at 7:24 AM, Igor <[email protected]> wrote:
> Dear Developers, > > Recent nvidia graphics chips have hundreds of cores. I also noticed > that recent versions of trilinos can be compiled with CUDA support. > Would CUDA capability of trilinos somehow affect fipy that is run in > parallel? > That's awesome news and maybe FiPy can leverage this capability. In all likelihood, FiPy also needs Numpy or a Numpy like replacement to be GPU capable as well. This is also a very important part of making FiPy GPU ready. If so, is there a way to run fipy scripts on multi-core gpu? > Someone has looked into using PyCUDA with FiPy, he recorded his thoughts on the wiki. http://matforge.org/fipy/blog/gpuarray http://matforge.org/fipy/blog/gpu_again I don't necessarily agree with the conclusions, but it is not as straightforward as simply using the correct back end tools. Given the required back end tools (Trilinos, PyCUDA), there are still a number of issues that will make this a challenge. How to minimize the communication between the CPU and the GPU is certainly one of those issues. If you do happen to try out a GPU ready version of Trilinos then let us know how it goes. Cheers, Daniel -- Daniel Wheeler
_______________________________________________ fipy mailing list [email protected] http://www.ctcms.nist.gov/fipy [ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ]
