ACCESS-OM2 Earth System Model Compatibility Between Two Supercomputers

The evolution of weather and climate modeling has been linked to advances in supercomputing, from the first atmospheric simulations on ENIAC in the 1950s to today’s exascale systems like El Capitan (1.7 exaflops). As Earth System Models (ESMs) have grown in complexity, integrating high-resolution atmosphere, ocean, ice, and biogeochemical components, they face critical challenges in scalability, energy efficiency, and portability across heterogeneous architectures. This study addresses these challenges through the porting of the ACCESS-OM2 ESM from Gadi (PBS, Intel Cascade Lake) to Leonardo (SLURM, Intel Sapphire Rapids) supercomputer, with a focus on workflow adaptation, cross-platform reproducibility, benchmarks and profiling. Scheduler adaptation included refactoring of the PBS and ACCESS OM2 Payu workflow manager for SLURM scheduler. ACCESS-OM2 profiling with “mpiP” and “perf” tools demonstrated the coupling between model components and I/O are the dominant bottlenecks and consume the most of the wall-clock model runtime. ACCESS-OM2 achieve weak scaling on Leonardo mostly due to not optimized NUMA usage in the ported configuration and requires further work to achieve high scalability. The study also contextualizes these results within broader ESM optimization efforts, drawing on case studies like CESM-HR on Sunway TaihuLight (3.4× speedup via cache-aware redesign) and EC-Earth3P-HR (46% runtime reduction through coupling synchronization fixes). The study also emphasizes the need for portable, scheduler-agnostic workflow managers and filesystems to match compute advancements that would support the global climate collaboration efforts, such as CMIP (Coupled Model Intercomparison Project).