Seismic Wave Propagation on GPUs

In this work, we present the adaptation of a 3D Finite Difference acoustic wave propagation<br>code to GPU architecture using the NVIDIA CUDA framework. We demonstrate the general<br>concept of transforming a CPU-based code to fit the requirements of GPU computing. The<br>individual steps of this process are illustrated along with the necessary modifications to the<br>original program code and algorithm.<br>Problems that typically arise when porting a code to graphics cards, such as numerical<br>stability due to reduced accuracy and memory availability/handling are discussed as well as<br>the ability to use multiple GPUs simultaneously.<br>The described technique results in a GPU-based code that provides speedups of one order of<br>magnitude in execution time compared to the original CPU version.