Current and emerging computing architectures require developers to cope with a high degree of per-node parallelism, new instruction sets, latency dominated strong scalability limitation and a heterogeneity of architectures, to name just a few. All these new challenges require disruptive developments, going beyond standard optimizations as they are typically used in HPC. It is in particular insufficient to treat this as a task of one single discipline, but to go interdisciplinary, involving deep understanding of HPC as well as mathematics.
Dr. Martin Schreiber will present and discuss various concepts and methods in this context, putting the focus on topics such as parallel- in-time methods (Rational approximations, Parallel Full Approximation Scheme in Space and Time), semi-Lagrangian time integration as well as machine learning.