[AstroPy] Image combine

[RayS]

FYI, I once tested pyCuda 
http://mathema.tician.de/software/pycuda  for 1D and 2D FFT 
computations. I found that the crossover on a multi-core Tyan MB was 
~4k N, below that, the (extra) bus transfer time exceeded the FFT speed-up.
But, I would expect the speed advantage for large images to be 
significant, particularly FITS cubes where multiple operations are 
performed while the data stays in GPU memory. With faster, wider 64 
bit bus break-even might be even lower - I should test on my new P6200 MB.

Ray