Accelerating 3-D Acoustic Full Waveform Inversion Using a Multi-GPU Cluster

Improving the computational efficiency of 3-D full waveform inversion (FWI) is a challenging task in seismic imaging. Using a multi-GPU cluster with an acceleration strategy to simulate wave propagation is an important means to improve its efficiency. We propose a multi-GPU acceleration 3-D acoustic FWI algorithm based on the finite-difference method in the time-domain (FDTD) method in this article. We improved the parallelism of the 3-D wavefield simulation algorithm based on a single GPU using a sliding 2-D thread block algorithm with three different 2-D shared memory stencils. For the multinode implementation, we achieved bidirectional parallel data transfer between GPUs and used multiple kernels to further overlap the calculation and transfer. Numerical tests verify the validity of our 3-D FWI algorithm accelerated with multi-GPU. The strategies used in our algorithm can significantly bring improvement in most cases. And the improvement is strongly related to the model size and the number of GPUs used. In our test, we achieve an acceleration of up to 19% in forward simulation and 25% in gradient calculation, compared with a typical multi-GPU implementation.

Automated summary

This article proposes a multi-GPU acceleration 3-D acoustic FWI algorithm based on the finite-difference method in the time-domain method. The algorithm improves the parallelism of the 3-D wavefield simulation algorithm and achieves bidirectional parallel data transfer between GPUs. Numerical tests show significant improvement in computational efficiency, with 19% acceleration in forward simulation and 25% acceleration in gradient calculation compared to a typical multi-GPU implementation.