It's not rare at all. A double takes up twice as much memory as a float and memory bandwidth is a precious commodity these days. If you have a lot of numbers (e.g. in 3D models or similar) and don't need the extra precision, using floats is an obvious choice.