You would probably get some benefit on a per CUDA core basis, but the extra space required would mean less CUDA cores, overall it would probably be a net loss o

You would probably get some benefit on a per CUDA core basis, but the extra space required would mean less CUDA cores, overall it would probably be a net loss on problems well suited to GPU acceleration