Starting with my conclusion GPU compute wins big time with tiles at 512*512 with the time of 01min28secs.
Following my test results:
Blender cycles render times
Blender version 2.79
renderer: GeForce GTX 780/PCIe/SSE2
vendor: NVIDIA Corporation
version: 4.6.0 NVIDIA 398.11
file BMW27.blend
Processor 3.40GHz, 4 Cores, 4 Logical Processors
Memory(RAM) 16,0 GB
Windows version 10
Renderer CPU
Tiles 32*32
Time 04:55 (mm:ss)
Renderer GPU (CUDA)
OpenSubdiv GLSL compute
Tiles 32*32
Time more than 15 minutes , I ended this render during rendering 
Renderer GPU (CUDA)
OpenSubdiv GLSL compute
Tiles 128*128
Time 02:15 (mm:ss)
Renderer GPU (CUDA)
OpenSubdiv GLSL Transform Feedback
Tiles 128*128
Time 02:11 (mm:ss)
Renderer CPU
Tiles 128*128
Time 05:02 (mm:ss)
Renderer GPU (CUDA)
OpenSubdiv GLSL compute
Tiles 256*256
Time 01:37 (mm:ss)
Renderer CPU
Tiles 256*256
Time 05:43 (mm:ss)
Renderer GPU (CUDA)
OpenSubdiv GLSL Transform Feedback
Tiles 512*512
Time 01:28 (mm:ss)
Renderer GPU (CUDA)
OpenSubdiv GLSL compute
Tiles 512*512
Time 01:28 (mm:ss)
Renderer CPU
Tiles 512*512
Time 11:01 (mm:ss)
Renderer GPU (CUDA)
OpenSubdiv GLSL compute
Tiles 960*540
Time 01:33 (mm:ss)
Renderer CPU
Tiles 960*540
Time 19:07 (mm:ss)