The scene file has now changed slightly and only 100spp. So new reference. (GTX1070 +Ryzen 1700X DDR4 @stock 2400mhz i think)
GPU+CPU 32x32 = 1:41
GPU +CPU 64x64 = 2:10
CUDA only at 64x64 =3:20
CUDA only 96x96 = 3:18
CUDA only 256X256 = 2:55
CUDA only 512X512 (and some smaller tiles to fit the gaps ) =2:51
CUDA+CPU 512x512 =(one eternity later) 16:13