Veris: fast & efficient sea-ice modeling in Python with GPU acceleration
- 1Alfred-Wegener-Institut, Helmholtz Zentrum für Polar- und Meeresforschung, Bremerhaven, Germany
- 2Niels Bohr Institute, University of Copenhagen, Copenhagen, Denmark
Abstract
Climate models have traditionally been developed in Fortran, due to its long-standing use in scientific computing and its excellent computational performance. While Python offers substantial advantages in terms of code readability, maintainability, and the availability of libraries and tools, the performance gap between Python and Fortran has historically limited Python's use in large-scale climate modeling. This performance gap can be mitigated by using the JAX library, which significantly improves execution speed of Python code. We use JAX as a backend for Veris, a new sea ice model implemented in Python. Veris builds upon the Fortran-based sea ice component of the general circulation model MITgcm. Benchmark experiments show that Veris exhibits scaling behavior with increasing process counts comparable to the Fortran reference implementation. For small CPU process counts, Veris outperforms the MITgcm, showing the great potential that JAX has for climate modeling, particularly as further improvements in inter-process communication are anticipated. When executed on a high-end GPU, a single-process Veris simulation matches the performance of the parallelized Fortran reference running on hundreds of CPU cores, but at a fraction of the energy cost. These results demonstrate the potential of Veris for large-scale HPC-based simulations.
Related
- Primula veris flower polymorphism at Sainte-Baume (Var, France)
- SamudrACE: Fast and Accurate Coupled Climate Modeling With 3D Ocean and Atmosphere Emulators
- An accurate and efficient framework for modeling multimetal competitive adsorption on clay minerals
- HydroCraft: an efficient online platform for hydrological modeling integrated with a large language model agent
- Development and validation of ARMS-gb v2.0: extending fast radiative transfer modeling capability to all-sky conditions for ground-based microwave radiometer retrievals
BibTeX
@article{gartner2026veris,
title = {{Veris: fast & efficient sea-ice modeling in Python with GPU acceleration}},
author = {Jan P. Gärtner and Martin Losch and Suvarchal K. Cheedela and Markus Jochum and Roman Nuterman},
journal = {Geoscientific Model Development},
year = {2026},
month = jun,
doi = {10.5194/gmd-19-5225-2026},
url = {https://doi.org/10.5194/gmd-19-5225-2026},
note = {CC BY},
}
RIS
TY - JOUR
TI - Veris: fast & efficient sea-ice modeling in Python with GPU acceleration
AU - Jan P. Gärtner
AU - Martin Losch
AU - Suvarchal K. Cheedela
AU - Markus Jochum
AU - Roman Nuterman
PY - 2026
DA - 2026/06/01
JO - Geoscientific Model Development
DO - 10.5194/gmd-19-5225-2026
UR - https://doi.org/10.5194/gmd-19-5225-2026
AB - Climate models have traditionally been developed in Fortran, due to its long-standing use in scientific computing and its excellent computational performance. While Python offers substantial advantages in terms of code readability, maintainability, and the availability of libraries and tools, the performance gap between Python and Fortran has historically limited Python's use in large-scale climate modeling. This performance gap can be mitigated by using the JAX library, which significantly improves execution speed of Python code. We use JAX as a backend for Veris, a new sea ice model implemented in Python. Veris builds upon the Fortran-based sea ice component of the general circulation model MITgcm. Benchmark experiments show that Veris exhibits scaling behavior with increasing process counts comparable to the Fortran reference implementation. For small CPU process counts, Veris outperforms the MITgcm, showing the great potential that JAX has for climate modeling, particularly as further improvements in inter-process communication are anticipated. When executed on a high-end GPU, a single-process Veris simulation matches the performance of the parallelized Fortran reference running on hundreds of CPU cores, but at a fraction of the energy cost. These results demonstrate the potential of Veris for large-scale HPC-based simulations.
SN - 1991-959X
SN - 1991-9603
ER -