I wouldn't advise either for fast operations but I would believe that a operations on floats would be faster as they are 32 bit vs 64 bit in doubles.