Metraipolis › archive › dc_fd5bae239079
Half-precision inference of AI weather prediction models: negligible loss of skill at a fraction of the cost — arXiv:2611.09573
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2611.09573
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-09T22:37:13.037Z
- SHA-256 of the archived text
1eb8bb6eeb00c383801fc131df393263efdbb3edae26e83965664b0143ef51b1- Stable URL of this page
https://metraipolis.com/documento/dc_fd5bae239079
How to cite
Half-precision inference of AI weather prediction models: negligible loss of skill at a fraction of the cost — arXiv:2611.09573 — arXiv (en), https://arxiv.org/abs/2611.09573 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_fd5bae239079 (Metraipolis archive dc_fd5bae239079, sha256 1eb8bb6eeb00c383801fc131df393263efdbb3edae26e83965664b0143ef51b1)
BibTeX
@misc{dcfd5bae239079,
title = {Half-precision inference of AI weather prediction models: negligible loss of skill at a fraction of the cost — arXiv:2611.09573},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2611.09573},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_fd5bae239079; sha256 1eb8bb6eeb00c383801fc131df393263efdbb3edae26e83965664b0143ef51b1},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.09573. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
AI weather prediction models are released with 32-bit (fp32) weights and run in fp32 by default, although current GPUs compute much faster in 16-bit formats. We run five publicly available models -- Aurora, Pangu-Weather, FourCastNet, SFNO and FastNet -- on a single 16 GB GPU from 52 initial dates in 2024, in fp32 and with their networks in half precision (fp16), and verify the forecasts against ERA5. The physical state must stay in fp32: converting the whole model rounds it in physical units, which in fp16 overflows and in bf16 discards meaningful information. With the state, its normalisation and denormalisation in fp32, running the network in fp16 changes the global RMSE by at most 1.4%, and by about 0.5% or less for four of the five models, equivalent to less than one hour of lead time up to day 8 and at most two hours at day 10. The differences show no systematic pattern by region, scale or for tropical and extratropical storms, apart from a small tropical penalty in FourCastNet, and fp16 leaves the CRPS, spread and calibration of Aurora's ensemble almost unchanged. No value overflowed fp16's numeric range (+/-65,504); the largest stay 38 to 544 times below it. A 10-day forecast in fp16 is 1.3-3.5 times faster than in fp32 and uses 1.4-3.7 times less energy. Sensitivity experiments show that rules of thumb from other applications of reduced precision carry over only partly: the first and last layers dominate for some models but not for others, and no type of layer is sensitive on its own. For running AI weather models on a single GPU, fp16 with the state in fp32 is practically lossless at a fraction of the cost. bf16, often preferred for its wider range, is as fast but less accurate than fp16 for every model, and simple 8-bit quantisation costs far more skill. Tema degli esploratori: calcolo e infrastrutture (elenco arXiv cs.PF, nuovi lavori del giorno). Categorie arXiv: Atmospheric and Oceanic Physics (physics.ao-ph) ; Machine Learning (cs.LG); Performance (cs.PF). Pagina del lavoro: https://arxiv.org/abs/2611.09573 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-11-09T22:31:47.217Z dalla pagina https://arxiv.org/list/cs.PF/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).