Yes, Float16 <http://en.wikipedia.org/wiki/Half-precision_floating-point_format> is explicitly intended for storage only:
It is intended for storage (of many floating-point values where higher > precision need not be stored), not for performing arithmetic computations. On Fri, May 8, 2015 at 9:26 AM, Sisyphuss <[email protected]> wrote: > I think since Float16 type will be converted to Float32 during a > calculation, there's no point to stock an array of Float16 (it will only > slow down the computation), unless the main memory is highly demanding. >
