Holzer M, Bauer M, Rüde U (2021)
Publication Language: English
Publication Type: Conference contribution, Abstract of a poster
Publication year: 2021
Event location: Online
We present a multiphase Lattice Boltzmann method based on the conservative Allen Cahn model. This approach is suitable for high density ratios and high Reynolds numbers. The code generation framework lbmpy is used to produce optimized code for CPUs and GPUs. A roofline analysis demonstrates the excellent efficiency of the generated GPU code. The resulting single-GPU code can be integrated into the multiphysics framework waLBerla to enable massively parallel simulations on very large domains. Scaling experiments are conducted on the PizDaint supercomputer up to 2048 GPUs. Near-perfect scaling behavior can be demonstrated when additional strategies are implemented to hide communication. The standard bubble rise scenario is used to validate the implementation. However, the parallel multi-GPU implementation enables simulations of much larger large scenarios with several hundred fully resolved bubbles.
APA:
Holzer, M., Bauer, M., & Rüde, U. (2021, July). lbmpy: Fast and Flexible Multi-Phase Lattice Boltzmann Simulations for High Density Ratios with Code Generation. Poster presentation at The Platform for Advanced Scientific Computing (PASC) Conference, Online.
MLA:
Holzer, Markus, Martin Bauer, and Ulrich Rüde. "lbmpy: Fast and Flexible Multi-Phase Lattice Boltzmann Simulations for High Density Ratios with Code Generation." Presented at The Platform for Advanced Scientific Computing (PASC) Conference, Online 2021.
BibTeX: Download