Domain decomposition method for 2d and 3d transport calculations using hybrid mpi/openmp parallelism

APOLLO3® and TRIPOLI-4® are registered trademark of CEAInternational audienceIn this paper we analyze the efficiency of the Domain Decomposition Method associated to the Coarse Mesh Finite Difference acceleration. We evaluate the effectiveness of the algorithm for shared memory parallelism. We also present the advantages of the Hybrid OpenMP/MPI parallelism to perform high-fidelity large scale calculation. We show that CPU time for a best-estimate 2D whole core calculation can be reduced from several days to few minutes on computer cluster. Finally high-fidelity 3D full assembly color-set calculation is compared to Monte Carlo simulation. This particular test shows the challenges for advanced neutron transport simulations in 3