9:45 AM - 10:00 AM
[2B01-06-04] Boosting an explicit and semi-implicit SPH code with dynamic load balancing parallelization on many GPUs
Keywords:Smoothed particle hydrodynamics, Graphics processing unit, Message passing interface, Dynamic load balancer, Large-scale simulations
The SPH method are attracting attention as an efficient method for solving free-surface flows in complex geometries such as tsunami run-up phenomena. In the case of urban-scale disaster analysis, the particle method must be applied to a large area with evenly spaced particles, which results in a large computational model. However, the memory size of GPUs is smaller than that of CPUs, and therefore, multiple GPUs are required to perform large-scale analysis. In this study, the parallel SPH particle computation using many GPUs including communication across computational nodes is developed.