Performance analysis of the free surface lattice Boltzmann implementation in waLBerla

Abstract The free‐surface lattice Boltzmann method uses a volume of fluid approach to simulate immiscible two‐fluid flow problems. It divides the simulation domain into three distinct phases—gas, fluid, and interface—where computation within the gas phase is disregarded. The interface delineates a one‐cell‐thick layer between the first two phases, validated physically for implementation in the HPC C++ multiphysics framework waLBerla but lacking an exhaustive performance analysis. This paper aims to shed light on node‐level performance on different architectures, employing continuous benchmarking, showing and analyzing weak scaling results on a modern HPC cluster, the Fritz supercomputer, and reporting energy consumption for the current implementation.