With screws, having removed the cooler of a graphics card myself I can tell you there are a lot of them. Look at the back of the card, there should be mounting points at the very least around where the GPU is (you can tell by a dense cluster of electronics).
The GPU is not designed to be interchangeable.
Not much really, ultimately they both crunch numbers. However the GPU is far more suited for parallel processing (Which is why they have a ridiculous number of cores) because thats what benefits rendering software best. They also have various other technologies like shader units.