Got take, but a double with only 8 bits of exponent actually seems kind of nice, you get the extra precision but it can be cast down to a single and you only lose precision; you don’t have values that are outside the range of singles.