It sums in floating point.
16 bit = not so good.
24 bit = better than 16.
32 bit = better than 24.
64 bit = better than 32.
32 bit float = better tan 64 bit??? Or not… It gets pretty academic at this point, not even Batman and Superman combined can hear the difference at this point.
32 bit float gives practically unlimited dynamics.
Anyway, there are looong debates on this on the web, google a bit.