3 ms·

Not all. In fact a small model that has none of them but loads them on demand might be the most efficient thing