Examining the GPU Acceleration Speed-Up for Finite Element Modeling of Additive Manufacturing

Betts, J. L., Downs, W., Dantin, M., & Priddy, M. W. (2022). Examining the GPU Acceleration Speed-Up for Finite Element Modeling of Additive Manufacturing. Solid Freeform Fabrication 2022: Proceedings of the 33rd Annual International Solid Freeform Fabrication Symposium. University of Texas - Austin.

Abstract

Using a graphics processing unit (GPU) in addition to a central processing unit (CPU) has demonstrated promise for the acceleration of processing-intensive operations such as finite element (FE) simulations. Commercial FE solvers have begun to utilize GPU acceleration for classical multi-physics applications, but the speed-up for additive manufacturing (AM) simulations is not well understood. There is a significant need for GPU acceleration for metal-based AM FE simulations, which are computationally expensive because of the high mesh densities and the large number of time increments employed. This study examines the efficacy of GPU acceleration for Abaqus AM simulations, where benchmark simulations using a sequentially coupled FE thermo- mechanical model are run both with and without GPU acceleration. The speed-up is compared across the AM process for the thermal and mechanical analysis. In this study, GPU acceleration provided the ability to decrease simulation runtime by two-to-four times on 4-8 CPU cores, and one-to-two times on 16-32 CPU cores