Parallelizing Lattice Boltzmann Simulations on Heterogeneous GPU–CPU Clusters
SESSION: Research Poster Reception
EVENT TYPE: Poster
TIME: 5:15PM - 7:00PM
AUTHOR(S):Johannes Habich, Christian Feichtinger, Georg Hager, Gerhard Wellein
ABSTRACT: Sustaining a large fraction of single-GPU performance in parallel computations is considered to be the major problem of GPU-based clusters.
We address this topic in the context of a lattice Boltzmann flow solver that is integrated in the WaLBerla software framework.
We show a multi-GPU implementation using a block-structured MPI parallelization, suitable for load balancing and heterogeneous
computations on CPUs and GPUs.
The influence of data transfer overhead on performance and scalability is studied in detail for weak and strong scaling scenarios. We also demonstrate how the need for CPU-GPU load balancing impacts performance.
Johannes Habich - Regional Computing Center Erlangen
Christian Feichtinger - University of Erlangen-Nuremberg
Georg Hager - Regional Computing Center Erlangen
Gerhard Wellein - Regional Computing Center Erlangen