4 ms·

But it is indeed quite a bit more computationally intendive, especially for the highest quality models.