The Potential Impact of GPUs on LS-DYNA Implicit

This talk will report the on-going efforts of LSTC to study the impact of GPUs on LS-DYNA. GPUs offer very high performance computational power at the cost of importing and exporting of data between the host computer and the GPU. The GPU has restricted memory, requires programming in a special language, and suffers performance reduction for double precision arithmetic. Still GPUs appear to offer a potential speed-up of a factor of 2 to 3. Initially our study is focusing on the impact on Implicit Mechanics. The most important computational kernel for Implicit Mechanics is in the direct solution of the sparse systems of linear equations. We will focus on how one can use GPUs for this computational kernel and the probable performance improvement.