Milestone M2.2: Reduced precision tools are usable within NEMO and IFS-FVM
This milestone successfully validates numerical precision reduction (half and bfloat16) in both the NEMO and PMAP climate models, demonstrating significant memory footprint reductions without compromising solution quality. Furthermore, all associated tools and mixed-precision codebases—including AutoRPE and GT4Py—have been made fully open-source to drive high-performance computing advancements across weather and climate modeling.
Authors
- Oriol Tintó Prims (ORCID: https://orcid.org/0000-0003-3182-3942)
- Peter Dominik Dueben (ORCID: https://orcid.org/0000-0002-4610-3326)
- Sara Faghih-Naini (ORCID: https://orcid.org/0009-0009-1880-2822)
- Maitane Fariñas (ORCID: https://orcid.org/0009-0003-4657-0834)
Institutions
- European Centre for Medium-Range Weather Forecasts (GB)
- Barcelona Supercomputing Center (ES)
- University of Oxford (GB)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-10-28
- DOI
- https://doi.org/10.5281/zenodo.23728825
- Primary Topic
- Parallel Computing and Optimization Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- European Commission