Interesting, my best time came from ignoring the ‘Powers of two’ rule. I feel like there might be some room for optimization external to blender. Something extracurricular for me to do some research on later.
My system:
CPU: AMD FX-8350 8-core @ 4GHz
GPU: Nvidia GTX 980
OS: Windows 10 1607 14393.187 (x64)
Tile Size: 960 x 540
Time: 2 min 3 seconds (GPU - CUDA)
Tile Size: 512 x 512
Time: 1 min 41 seconds (GPU - CUDA)
Tile Size: 480 x 270
Time: 1 min 36 seconds (GPU - CUDA)
Tile Size: 240 x 135
Time: 1 min 48 seconds (GPU - CUDA)
Tile Size: 128 x 128
Time: 1 min 52 seconds (GPU - CUDA)
Time: 9 min 51 sec (CPU)
Tile Size: 64 x 64
Time: 3 min 34 seconds (GPU - CUDA)
Time: 9 min 17 sec (CPU)
Tile Size: 8 x 8
Time: 8 min 48 sec (CPU)