It should be possible to split big volume into smaller chunks, do transformation on GPU and then combine the chunks on CPU