High Performance Computing
Our code has been developed with massive parallelisation on all levels in mind, which can be challenging given the large, looming communication overhead that the radiative transfer problem might bring about if not properly disentangled.
PHOENIX has been adapted to benefit from a host of different hardware architectures. It, has proven to run reliably with tens of thousand of parallel tasks and made use of infrastructure such as GPUs, GPGPUs, Intel Xeon Phi (1st and 2nd generation) and more exotic equipment. Depending on available hardware, the code is designed to be easily customized to make the best of available ressources. Parallelisation is mediated via MPI, OpenMP, and OpenCL, depending of the selected configuration.
Special attention is devoted to ensure that results are not affected by the choice of hardware or employed compilers.