Utilizing Intel Xeon Phi Coprocessors Concurrently with Intel Xeon Processors to Accelerate WRF Simulation Throughput

Since the release of version 3.5, WRF is now able to run on Intel Xeon Phi coprocessors. Currently, the only parameterization scheme optimized to be fully vectorized for the Xeon Phi is the WSM5 microphysics scheme in WRF. Despite this fact, WRF still performs well when run natively on the Xeon Phi coprocessors. To further accelerate research, a novel approach can be taken to increase throughput where a large number simulations must be completed. In this regard, two separate instances of WRF may be executed concurrently, one on the coprocessors and another running alongside on the host processors. This work will demonstrate the procedure and performance of WRF run in this manner on the Purdue University Conte cluster in addition to the benefits afforded to a current research project.