Mathematica uses arbitrary-precision to represent numbers that would be denormalized:
This conversion can be prevented in
Compile and some other functions with a system option:
The smallest denormalized machine number:
Restore the option to its default: