2 ms·

Could there be a similar scheme for floats? Also, is there a penalty for using float16s for CPU centric code?

• akrasuski1 9y ago
  Using fixed-point representation would allow using this compression method.