Converting a floating-point value to an integral value removes the fractional part, provided the result can be represented in the integral type (i.e. the value isn't too large to fit). Inserting into a stream rounds the value, which can produce a different result.