nekCRF: A GPU-Accelerated Spectral Element Reactive Flow Solver
Monday, May 13, 2024 3:00 PM to Wednesday, May 15, 2024 4:00 PM · 2 days 1 hr. (Europe/Berlin)
Foyer D-G - 2nd floor
Research Poster
Computational PhysicsEngineeringExtreme-scale AlgorithmsMixed Precision
Information
Poster is on display and will be presented at the poster pitch session.
Most of the performance of the latest generation of HPC systems, and especially all exascale supercomputers, is based on GPUs. Many codes were originally designed to run on CPUs, and switching to GPUs is highly non-trivial. This is especially true for reactive flow solvers due to the large number of scalar fields and the resulting stiff systems. However, these are absolutely required for accelerating the green transformation of the society. In the frame of the Center of Excellence in Combustion (CoEC), the development of nekCRF, a GPU-based solver for extreme-scale computational fluid dynamics (CFD) for low Mach number compressible reactive flows has been enabled. nekCRF is based on the well-established high-order spectral element code nekRS and solves the low-Mach form of the Navier-Stokes equations together with the energy and species transport equations. nekCRF was run on up to 3600 NVIDIA A100 GPUs of JUWELS Booster, the fastest German supercomputer, and showed excellent scaling. Single node performance was demonstrated to be very efficient due to code design and high-performance computing (HPC) optimizations. Several test cases, ranging from canonical tests and flame kernels in confined geometries to complex applications with moving meshes, confirmed the numerical and physical accuracy of nekCRF. Comparisons with competing reactive flow solvers show significant speedup by nekCRF. Overall, nekCRF is fully exascale-ready and can be used to accurately simulate turbulent flames in open or closed domains with unprecedented resolution.
Contributors:
Most of the performance of the latest generation of HPC systems, and especially all exascale supercomputers, is based on GPUs. Many codes were originally designed to run on CPUs, and switching to GPUs is highly non-trivial. This is especially true for reactive flow solvers due to the large number of scalar fields and the resulting stiff systems. However, these are absolutely required for accelerating the green transformation of the society. In the frame of the Center of Excellence in Combustion (CoEC), the development of nekCRF, a GPU-based solver for extreme-scale computational fluid dynamics (CFD) for low Mach number compressible reactive flows has been enabled. nekCRF is based on the well-established high-order spectral element code nekRS and solves the low-Mach form of the Navier-Stokes equations together with the energy and species transport equations. nekCRF was run on up to 3600 NVIDIA A100 GPUs of JUWELS Booster, the fastest German supercomputer, and showed excellent scaling. Single node performance was demonstrated to be very efficient due to code design and high-performance computing (HPC) optimizations. Several test cases, ranging from canonical tests and flame kernels in confined geometries to complex applications with moving meshes, confirmed the numerical and physical accuracy of nekCRF. Comparisons with competing reactive flow solvers show significant speedup by nekCRF. Overall, nekCRF is fully exascale-ready and can be used to accurately simulate turbulent flames in open or closed domains with unprecedented resolution.
Contributors:
Format
On-site
