Eulerian incompressible smoothed particle hydrodynamics on multiple GPUs

Recent advances in the development of Eulerian incompressible smoothed particle hydrodynamics (EISPH), such as high-order convergence and natural coupling with Lagrangian formulations, demonstrate its potential a meshless alternative to traditional computational fluid dynamics (CFD) methods. This work aims address one major outstanding limitations EISPH, relatively high cost, by providing an implementation that can be deployed on multiple graphics processing units (GPUs). To this end, pre-existing multi-GPU version open-source weakly-compressible code DualSPHysics is converted EISPH formulation integrated multigrid solver (AmgX) treat pressure Poisson equation implicitly. The integration AmgX within presents significant challenge, since designed for distributed systems therefore conflicts single-node shared memory design code. present validated against well-known test cases, showing excellent agreement benchmark solutions demonstrating second-order convergence. A detailed profiling performance testing also presented investigate consumption scaling characteristics. results show approximately 87%–95% strong efficiency 92%–94% weak both 2D 3D up four GPUs. Large spikes during initialisation linear library are found impede full utilisation device memory. Nevertheless, shown permit problem sizes order 69.3 million (2D) 20.8 (3D) particles GPUs (128 GB total memory), which beyond what has previously been reported SPH GPUs, where PPE treated implicitly, demonstrates CFD