3.2 MPAS on GPUs.
Suresh, Supreeth, NCAR/UCAR, and Raghuraj Kumar, NVIDIA
The Model for
Prediction Across Scales - Atmosphere (MPAS-A) is a general circulation
(global) model of the Earth's atmosphere that is designed to work down to
non-hydrostatic scales where convective (vertical) cloud processes are
resolved. To date, MPAS-A has been used primarily for meteorological research
applications, although climate applications in the community earth system model
are being contemplated.
The NCAR's Weather and Climate Alliance (WACA) project is focused on
developing a performance portable, implementation of MPAS-A for both Central
Processing Units (CPUs) and Graphics Processing Units (GPUs) using the OpenACC
directive-based parallelization language. The WACA project team consists of
NCAR staff, faculty and students at the University of Wyoming, NVIDIA &
NVIDIA PGI developers, and has received the collaboration and support of The
Weather Company and IBM, Inc., and the Korea Institute for Science and
Technology Information (KISTI). This talk presents MPAS-A performance results,
the status of current work and future plans for enabling MPAS-A execution on
GPUs.