It doesn't really matter HOW they do it, as long as you get the advantages of float encoding (practically infinite headroom). Of course if you zoom in enough there will be something in there that uses integers, but this would be true for e.g. a floating point adder as well.