Achieving 1.88x Speedup in WRF with OpenMP Offload and Codee

For those working with simulation software, here’s a deep dive into our recent collaboration with NERSC of the U.S. Department of Energy. Together, we optimized WRF’s performance on the Perlmutter supercomputer at NERSC using Codee and OpenMP offload.

The work focused on speeding up Weather Research and Forecasting Model’s Fast Spectral Bin Microphysics (FSBM) routine by offloading computations to GPUs. With the support of Codee’s static code analysis, this approach achieved a 1.88x speedup in the CONUS-12km winter storm test case, significantly boosting efficiency and forecast accuracy.

:link: Read the full success story poster here for detailed methods and results: https://docs.nersc.gov/applications/wrf/files/WRF_GPU_poster.pdf

We’ll be presenting this work at SC24’s WACCPD workshop. Join us on Monday, Nov 18th, at 5 pm EST to learn more!

10 Likes