Abstract of GBSHYB Project

Cluster systems built with general-purpose processors are superseded today by platforms built around a hybrid architecture, with nodes constituted by processors based on CPU cores and many-core accelerated devices (MIC) or graphical processing units (GPU). By being able to exploit the capabilities of these computers, we expect to be able to significantly enhance the computational capabilities of codes currently being executed on standard platforms. We therefore propose to carry out the necessary steps to port the GBS code, used to simulate plasma turbulence in the tokamak SOL, to hybrid architecture computers. In particular a mixed communication (OpenMP+MPI+OpenACC or Intel Offload OpenMP) will be implemented in GBS, first focusing on the parallel multigrid solver recently implemented in GBS. GBS will then be tested on the supercomputers based on hybrid architecture. By being able to exploit these computers, we expect GBS to perform full size simulations of SOL plasma turbulence in JET and ITER that are presently out of reach.