I guess with GPU, we can mimic the lens system by changing pinhole model to a model where ray shoot to a square randomly. In this way, we can get the depth of field feeling as it is in DSLR.