Optical flow settings

I notice the macros using optical flow usually specify the use of 32 bit float. Does this give better accuracy if the vectors are used on the fly or is it of benefit only if the vectors are saved to exr files and reloaded?