Would it be possible to run the iterations in parallel with the GPU and CPU for the same input to improve time efficiency?