Using MIC cards

MIC accelerators, like GPUs, have enormous floating point performance and memory bandwidth.  Each card (chip) contains 60+ x86 cores, and can be programmed in C or C++, and code can be cross compiled for the MIC using the standard Intel compiler.

To develop for MIC, users can login to the node qcd12kmi. This node has one Intel Xeon Phi accelerator. Users can test their code on this accelerator to verify programs working as expected on the accelerator.

Programming Model

MIC cards can be used as offload accelerators, just as GPUs are used.  In this mode, the offloaded routine is separately cross compiled.

Alternatively, the MIC nodes can run entire applications as a standalone job or as a MPI process (rank).

The following sections cover how to run with MIC accelerators.