Improving LSTC’s Multifrontal Linear Solver

This paper is no longer available. You will be redirected to more recent and relevant content.

Improving the performance of LS-DYNA’s multifrontal solver can result in near-peak throughput on computations performed regularly today on multicore clusters. Because of the solver's importance to LS-DYNA, LSTC is continuously researching how to improve it. In this project, experts from LSTC collaborated with Cray and Intel, using a Rolls Royce engine model to perform studies on reducing the bottleneck for implicit calculations.

In this paper, you’ll learn about:

The impact of changes in microprocessor architecture and new technology for SSDs

Ways to substantially decrease memory footprint of factorization 

The dramatic impact on linear solver performance when operating out-of-core

Paper to be presented at the 2017 European LS-DYNA Conference by:

  • Roger Grimes, Robert Lucas, Francois-Henry Rouet and Clement Weisbecker, LSTC
  • Nick Meng, Intel Corp.
  • Ting-Ting Zhu, Cray Inc.