Answers are grounded in this paper and cite their sources.
Fusion research and the design of future fusion reactors require models capable of reproducing complex plasma behaviour and managing coupled multiphysics workflows, which depend on large high-performance computing (HPC) resources. However, many established fusion codes were originally developed for CPU-based systems and face challenges in exploiting heterogeneous HPC architectures, including difficulties in Graphics Processing Unit (GPU) utilization, scalability limitations, insufficient software portability, and maintainability. In this context, EUROfusion established Advanced Computing Hubs (ACHs) to support scientists in adapting their codes to current supercomputers. This paper reviews recent work of the CIEMAT-BSC ACH through four representative fusion codes: BIT1, JOREK, Alya/NEUTRO, and SOLPS-ITER. The activities include performance profiling, optimization, GPU porting, solver library evaluation, memory analysis, numerical validation, and energy-efficiency assessment. The results include the successful GPU porting of BIT1, with most of the computational workload moved to the GPU; up to 3× faster execution of ELPA2 than ScaLAPACK in selected CPU benchmarks; a more than 2.6× reduction in energy consumption for Alya/NEUTRO; and up to a 39% runtime reduction for SOLPS-ITER. These cases illustrate practical challenges and lessons learned when modernizing fusion applications for current HPC systems. The paper also discusses the role ACHs could play in the future.
No takes yet. Share an insight, caveat, or question.
Sáez et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: