This is what I found as well, and allude to in other comments in the thread. Particularly when you look at GPGPUs like the NVidia Tesla, they are generally pretty terrible when priced per-gflop because they live in a niche market that wants to handle finite element analysis and supercomputing tasks.

Image processing doesn't require double precision, so we don't need GPUs tuned for it, which means we can use Fire Pro's and similar workstation or server grade cards.