Toward GPU-Resident Climate Models: A Feasibility Study on Lossy Compression for the Spherical Harmonic Transform's Communication Bottleneck
Abstract
Operational pseudospectral atmospheric models such as the ECMWF Integrated Forecasting System (IFS) run today almost exclusively on CPUs; GPU ports are under active development but not yet used in production. These models rely on the Spherical Harmonic Transform (SHT). Each time-step requires forward and inverse SHTs, and both passes depend on global pencil transposition that redistribute multi-dimensional arrays across compute nodes. At large node counts these global collectives dominate wall-clock time. We investigate GPU-resident lossy compression, using representative fields from the DYAMOND high-resolution operational dataset as input, and combining measured GPU compression throughput with SimGrid network simulation, we show that ZFP at 16 bits per value (rate-16), about the same storage budget as float16, matches the communication-time reduction of float16 truncation while delivering approximately 1600× lower mean relative error. ZFP at 8 bits per value (rate-8) achieves approximately 1.93× the speedup of float16 while retaining 4× lower mean relative error.
Related
- Transformation product bottleneck: Photo-recalcitrance of rotenolone limits the aquatic degradation of rotenone in a temperature-dependent manner
- The magmatic and tectonic role in spreading styles at ridge–transform intersections between the Marathon and Vema Transform Faults
- Concave sparsity-assisted time-reassigned multisynchrosqueezing transform for early-stage weak fault feature enhancement with applications to bearing fault diagnosis
- Spherical Physics-informed Neural Operator with Multi-scale Coupling for Meteorological Downscaling
- Ground roll suppression based on the synchrosqueezed wavelet transform and local adaptive subtraction filtering
BibTeX
@misc{breschi2026toward,
title = {{Toward GPU-Resident Climate Models: A Feasibility Study on Lossy Compression for the Spherical Harmonic Transform's Communication Bottleneck}},
author = {Lorenzo Breschi and Flavio Vella},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.24294},
eprint = {2609.24294},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.24294},
}
RIS
TY - UNPB
TI - Toward GPU-Resident Climate Models: A Feasibility Study on Lossy Compression for the Spherical Harmonic Transform's Communication Bottleneck
AU - Lorenzo Breschi
AU - Flavio Vella
PY - 2026
DA - 2026/09/21
PB - arXiv
DO - 10.48550/arxiv.2609.24294
UR - https://arxiv.org/abs/2609.24294
AB - Operational pseudospectral atmospheric models such as the ECMWF Integrated Forecasting System (IFS) run today almost exclusively on CPUs; GPU ports are under active development but not yet used in production. These models rely on the Spherical Harmonic Transform (SHT). Each time-step requires forward and inverse SHTs, and both passes depend on global pencil transposition that redistribute multi-dimensional arrays across compute nodes. At large node counts these global collectives dominate wall-clock time. We investigate GPU-resident lossy compression, using representative fields from the DYAMOND high-resolution operational dataset as input, and combining measured GPU compression throughput with SimGrid network simulation, we show that ZFP at 16 bits per value (rate-16), about the same storage budget as float16, matches the communication-time reduction of float16 truncation while delivering approximately 1600× lower mean relative error. ZFP at 8 bits per value (rate-8) achieves approximately 1.93× the speedup of float16 while retaining 4× lower mean relative error.
ER -