I ran my tests all using the default 1920 x 1080px at 50%. Times are approximate…
CPU
512 tiles, 12 mins
16 tiles, 5 mins
GPU
512 tiles, 8+ mins
256 tiles, 4 mins (winner!)
128 tiles, 5 mins
Looks like my system renders best using the GPU at 256 tiles.
I’m using a Radeon R9 200 series that does not have CUDA support (time to go shopping!)