3.3      GPU performance study for the WRF model on the Summit supercomputer.

Adie, Jeff, NVIDIA, Gökhan Sever, Rajeev Jain, DOE Argonne NL, and Aide, Jeffrey, NVIDIA

This presentation will examine HPC progress of GPU developments for the WRF model using OpenACC directives. GPU WRF development is based on ARW 3.7.1, and led by NVIDIA in collaboration with NCAR MMM. The implementation includes the complete dynamical core and a total of 10 physics options from 5 physics types, such that a full WRF model can be executed on GPU-based systems. Scientists at the US DOE Argonne National Laboratory, led a study to measure performance and scalability of GPU WRF on the Summit supercomputer. Results for science-scale simulations demonstrate substantial GPU speedups over CPU-only computing.