Domanda di colloquio di NVIDIA

How to process an image in CUDA? How to map to threads? Is it memory-bound or compute-bound?