Running with the CUDA backend