Aim to improve the performance of the current implementation, expand the task types that can run on the GPU, start looking towards MPI + GPU