One example is the Titan supercomputer (ranged as the 5th fastest in the world) which conatains 18,688 CPUs and 18,688 GPUs. The GPU involved is the Nvidia Tesla K20X GPU. This is a so-called General Purpose GPU (GPGPU). It doesn't contain a display output (later Tesla units do).
As far as I know (correct me if I'm wrong), the GPUs operate by transforming data into "image data".
Titan is used for scientific computing, which involve high-performance and very demanding simulations. You can read about it here:
https://en.wikipedia.org/wiki/Titan_(supercomputer).
pibbur who, having millions of image processors, thinks (hopes) that he still outperforms supercomputers when it comes to general image processing (especially segmentation and recognition).