Thanks for teaching me about changing the tile sizes, it makes a huge difference between GPU and CPU (which before this I ran at 64x64 for both).
My times:
CPU: Intel i7 5930K 3.5Ghz (6core, 12threads?)
GPU: Nvidia GTX Titan Black
OS: Windows 10 64bit
GPU(512x512): 01:06:97
GPU(256x256): 01:11:95
GPU(128x128): 01:39:xx
CPU(32x32): 02:50:77
CPU(16x16): 02:50:07