https://hgpu.org/?p=26471
Machine Learning for CUDA+MPI Design Rules