MORE: talk by Peter Bastian (Interdisciplinary Center for Scientific Computing, Heidelberg University)

Unleashing the high-performance advertised by current supercomputers provides quite a challenge for finite element methods. The traditional approach of setting up a stiffness  matrix in a sparse matrix format and solving it by some preconditioned Krylov space method performs typically only at a few percent of theoretical peak due to memory bandwidth limitations. Matrix-free implementation of high-order methods offers the possibility bypassing the memory bottleneck while at the same time reducing the number of operations  substantially. In this talk we present the sum factorization method applied to discontinuous Galerkin discretizations, present performance results for a convection-diffusion operator including a mostly matrix-free algebraic multigrid preconditioner, show applications to the incompressible Navier-Stokes equations and present first results of a code generation framework. The results presented in this talk are joint work with Steffen Müthing, Dominic Kempf, Marian Piatkowski and Eike Mueller.