Engineer want to use more resources to do finite element analysis including GPU. My question is if current Intel MPI can support GPU computing? If support, which options should be used when mpiexec or mpirun?
Appreciate for your informaiton.
MPI does not support GPGPU programming, which means you cannot offload ranks/processes on GPU as of now. Intel Offers a GPU language dpcpp as a part of OneAPI, using which you can code for heterogeneous platforms. If you have a multi-cluster environment you can use a Hybrid-Programming model which uses both MPI and DPCPP. You can launch one process/rank for each node and use dpcpp to spawn threads and accelerate the code using GPU on that node.
To get started with the OneAPI you can download OneAPI Basekit and HPC kit from here - https://software.intel.com/content/www/us/en/develop/tools/oneapi/base-toolkit.html
A sample DPCPP/MPI code - https://github.com/intel/HPCKit-code-samples/tree/master/MPI/mpi_dpcpp
Reach out to us if you need any further help.
We are closing this thread assuming your question has been answered.
If you require additional assistance from Intel, please start a new thread. Any further interaction in this thread will be considered community only.