Performance analysis of a hybrid MPI/CUDA implementation of the NASLU benchmark