PNNL

Abstract

In nature, many complex multi-physics coupling problems exhibit strong diffusivity inhomogeneity, where one process occurs several orders of magnitude faster than others. Simulating rapid diffusion alongside slower processes demands intensive computational resources due to the need for small time steps. To address these computational challenges, we have developed an efficient numerical solver named the Finite Difference informed Random Walker (FDiRW) solver. In this study, we propose a GPU-accelerated mixed-precision configuration for the FDiRW solver to maximize efficiency through GPU (Graphics Processing Unit) multi-thread parallel computation and lower precision computation. Numerical results reveal that the proposed GPU-accelerated mixed-precision FDiRW solver maintains numerical accuracy while achieving a 117x speedup through GPU acceleration, with an additional 1.75x speedup by employing lower precision GPU computation. Notably, for larger model sizes, the GPU-accelerated mixed-precision FDiRW solver exhibits a computational complexity of O(N_L), where N_L represents the number of nodes used in the simulations. For a medium-sized model of 192×192×192, this approach reduces the total computational time to 10 minutes, enabling the simulation of larger systems with strongly inhomogeneous diffusivity.