That is correct but if you do it inloop you need additional GPU resources at the time of the render that you can save out of loop for the overall Scene complexity.
Thus why some go the Post way of Denoising and overall in case of Nvidias AI Denoise which works very simple on the Final Output itself spatially and no other 3D layer via IRay neither spatial nor temporal bellow it, it shouldn't matter much for your final result at all.
But you will be able to render a more complex spatial scene with those saved resources at render time and avoid a possible CPU fallback of IRay on a more complex scene Setup giving you a bit more headroom when you need it (mostly on the edge of your physical usable system resources).