Fast and power efficient GPU-based explicit elastic wave propagation analysis by low-ordered orthogonal voxel finite element with INT8 Tensor Cores

Abstract

There is a strong need for faster and more power-efficient explicit elastic wavefield simulations for large and complex three-dimensional media using a structured finite element method. Such wavefield simulations are suitable for GPUs, which have been improving their computational performance in recent years, and the use of GPUs is expected to speed up such simulations. However, there is still room for speedup and improving energy efficiency of such simulations using GPUs, since the performance of GPUs is not fully exploited just by its simple use, and the conventional method involves some numerical dispersion. In this paper, we propose a method for fast and efficient explicit structured-mesh wavefield simulation on GPUs by utilizing INT8 Tensor Cores and reducing numerical dispersion. We implemented the proposed method on GPUs and evaluated its performance in detail using an application example that simulates a real problem, and showed that it is faster and more efficient than conventional methods on many-node CPU-based systems and multiple GPU-based systems. This paper is the extended version of Ichimura et al. (2024).[1]