I don’t know about you, but I typically have a number of “back-burner” projects on the go. Currently I'm playing with creating my own simple binary floating-point format as part of an educational tool ...

The new Half type is composed of 16 bits and will be geared towards speeding up machine learning workflows by enabling faster computation and smaller storage requirements at the expense of precision.

[Editor's note: For an intro to fixed-point math, see Fixed-Point DSP and Algorithm Implementation. For a comparison of fixed- and floating-point hardware, see Fixed vs. floating point: a surprisingly ...