Like many other people have already mentioned before me in this thread, it is also a matter of parallelization. You can imagine for yourself that key frames from a video stream can be compressed independently from one another so the high number of GPU shaders can work concurrently at full capacity.