Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
This seminar from the FEM@LLNL Series features Stefan Turek from Technical University Dortmund discussing hardware-oriented numerical methods for massively parallel and low precision accelerator hardware in computational fluid dynamics (CFD). Explore how modern High Performance Computing (HPC) facilities with millions of cores and fast but lower precision accelerator hardware can be leveraged through specialized numerical techniques for PDEs to achieve exceptional computational efficiency. The presentation focuses on nonstationary flow simulations involving hundreds of millions to billions of spatial unknowns across thousands or millions of time steps. Discover innovative "parallel-in-space global-in-time" Newton-Krylov Multigrid approaches that enable greater parallelism for exascale computing, and learn about the "prehandling" concept for transforming ill-conditioned equation systems to work effectively with lower precision hardware like GPUs. The talk includes numerical results demonstrating these concepts and discusses challenges specific to large-scale flow problems. Part of the MFEM project's seminar series on finite element research and applications.
