Can Wind Energy Simulations Reach Exascale?

The pursuit of exascale computing has far-reaching implications for scientists seeking to simulate complex phenomena like atmospheric boundary layer flows. In this article, we delve into the challenges and opportunities that arise when scaling up large-eddy simulation models on advanced computing architectures. The importance of accurately modeling these flows cannot be overstated, as they influence everything from transportation systems to renewable energy generation. By exploring the performance of open-source codes NekRS and AMRWind on exascale systems, researchers can develop more accurate and efficient simulations, ultimately contributing to a more sustainable energy future.

Can Wind Energy Simulations Reach Exascale?

The pursuit of exascale computing has been a long-standing goal for scientists and researchers, particularly in the field of wind energy simulations. In this article, we explore the challenges and opportunities that arise when attempting to scale up large-eddy simulation (LES) models for atmospheric boundary layer (ABL) flows on advanced computing architectures.

The Importance of ABL Flows

Atmospheric boundary layer flows are a crucial component of everyday life, influencing various aspects such as vertical exchanges in moisture, aerosols, and atmospheric gases. These flows also affect practical aspects like transportation systems, renewable energy generation, pollution dispersion, noise propagation, and transmission of electromagnetic signals. The complexity of ABL flows arises from factors like density stratification, Coriolis effects, regional-scale weather patterns, and terrain.

Large-Eddy Simulation (LES) for ABL Flows

LES is a numerical technique used to simulate ABL flows by solving the governing physics equations in filtered form. This approach allows for the direct resolution of larger energy-containing eddies while modeling smaller scales using subgrid-scale models. LES has been widely applied to study ABL flows, with significant research efforts focused on improving model fidelity and performance.

NekRS and AMRWind: Two Open-Source Codes

This study examines two open-source computational fluid dynamics (CFD) codes for simulating ABL flows: NekRS and AMRWind. NekRS is a high-order unstructured-grid spectral element code, while AMRWind is a block-structured second-order finite-volume code with adaptive mesh refinement capabilities.

Code Development and Performance

The objective of this study is to co-develop these codes to improve model fidelity and performance for ABL-based applications like wind farm analysis on advanced computing architectures. To achieve this goal, the researchers investigated the performance of NekRS and AMRWind on the Oak Ridge Leadership Facility supercomputers Summit using 4-800 nodes, 24-4800 NVIDIA V100 GPUs, and Crusher, the testbed for the Frontier exascale system using 18-384 Graphics Compute Dies on AMD MI250X GPUs.

Strong and Weak Scaling Capabilities

The study compared strong and weak scaling capabilities of NekRS and AMRWind, as well as linear solver performance and time to solution. The results showed that both codes exhibited good strong scaling up to 800 nodes, but weak scaling was more challenging. The researchers identified leading inhibitors to parallel scaling, highlighting the need for further optimization.

Conclusion

The pursuit of exascale computing for wind energy simulations is a critical step towards improving model fidelity and performance for ABL-based applications. This study demonstrates the potential of open-source codes like NekRS and AMRWind in achieving this goal. By understanding the challenges and opportunities that arise when scaling up LES models, researchers can develop more accurate and efficient simulations, ultimately contributing to the advancement of wind energy production.