The HiDALGO2 Centre of Excellence is pleased to announce a new publication by its researchers, focusing on assessing the impact of processor architecture on the performance of Computational Fluid Dynamics (CFD) applications.
The paper, titled “Evaluating AMD EPYC CPU architectures on CFD applications,” appears in Future Generation Computer Systems, Volume 177.
Key Insights from the Study
The research provides a comprehensive empirical analysis of how different generations of processor architectures influence the efficiency of CFD simulations, which are critical to the environmental use cases addressed by HiDALGO2,.
Among its key findings:
- Cross-Generational Analysis: The study evaluates a wide range of AMD EPYC processor generations, including Rome, Milan, Milan X, Genoa, Genoa X, and the novel Bergamo architecture. These were analyzed based on varying core counts (64-128), L3 cache sizes, and RAM types (DDR4 vs. DDR5).
- Real-World Application Models: The research was conducted using the OpenFOAM application, utilizing two specific memory-bound models: motorBike and the Urban Air Pollution model,. This directly supports the HiDALGO2 Urban Air Project (UAP) pilot, which relies on high-resolution CFD simulations for city airflow.
- New Performance Metric: The authors introduced the FVOPS (Finite VOlumes solved Per Second) metric. This innovation allows for a direct and accurate comparison of application performance across different hardware architectures.
- Hardware Interaction: Particular attention was paid to the effective use of L2 and L3 cache memory and memory channel bandwidth. Using profiling tools like AMD uProf and LIKWID, the researchers identified specific behaviors and potential inefficiencies in how memory-bound applications interact with the CPU.
Advancing Co-Design for Exascale
This study aligns with HiDALGO2’s mission to improve the scalability and efficiency of Global Challenges applications on High-Performance Computing (HPC) infrastructures.
By understanding the technological advancements of modern HPC systems, specifically the impact of large cache capacities and memory bandwidth, the project fosters a co-design approach. This ensures that software is adapted to better utilize available infrastructure features, paving the way for efficient simulations on pre-exascale and exascale systems.
Citation:
Lawenda, M., Szustak, Ł., Környei, L., Galeazzo, F.C.C., & Bratek, P. (2025). Evaluating AMD EPYC CPU architectures on CFD applications. Future Generation Computer Systems, 177, 108237. https://doi.org/10.1016/j.future.2025.108237
