The gist is they store uniformly distributed values between 0..1 or -1..1, which can be more space efficient than floats if you know your data will always be in

The gist is they store uniformly distributed values between 0..1 or -1..1, which can be more space efficient than floats if you know your data will always be in one of those ranges. The conversion to/from floats isn't too hard to do in software but it's common enough in graphics that GPUs usually have dedicated hardware support for it.