.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
P16 - GPU Porting of ECMWF Physical Parametrizations Using a High-Level Programming Model
Description
We present recent developments in the GPU porting using the domain-specific library GridTools for Python (GT4Py) of three physical parametrizations from the Integrated Forecasting System (IFS) of the European Centre for Medium-Range Weather Forecasts (ECMWF): the cloud microphysics packages CLOUDSC & CLOUDSC2, the radiation scheme ecRad, and the surface parametrization ecLand. We outline our high-level porting strategy of legacy Fortran physics codes, highlight performance results for CLOUDSC & CLOUDSC2, and discuss ongoing work with ecRad and ecLand.
Presenter(s)
Presenter
Gabriel Vollenweider studied environmental science at ETH Zürich, obtaining his Bachelor of Science in 2020 and his Master of Science in 2023. After ending 2023 with an internship in the scientific computing team at MeteoSwiss, he started a doctorate at ETH Zürich in 2024. In his doctorate, he is first implementing the required physical parametrizations in the Portable Model for Multi-Scale Atmospheric Prediction (PMAP; jointly developed at ECMWF, ETH Zürich, and CSCS), and later focuses on the evaluation of the coupling between the physical parametrizations and the finite-volume dynamical core.