A 3D radiative transfer framework: XII. Many-core, vector and GPU methods

3D detailed radiative transfer is computationally taxing, since the solution of equation involves traversing six dimensional phase space domain. With modern supercomputers hardware available for wallclock speedup rapidly changing, mostly in response to requirements minimize cost electrical power. Given variety computing architectures, we aim develop and adapt algorithms different architectures improve performance on a wide platforms. We implemented main time consuming kernels solving problems vastly using MPI, OpenMP, OpenACC vector algorithms. Adapted lead massively improved speed all making extremely large model calculations easily feasible. These would have previously been considered impossible or prohibitively expensive. Efficient use devices entirely feasible, but unfortunately requires implementation specialized them.