Since we now have sub-millisecond precision in the data, we control output
width based on the fractional part so we don't lose precision