there's no real reason to try and hit 0 offset anyways. if you get it stable, it usually implies you could just increase the multiplier and offset by some n>0 for greater overall performance.
The problem is that real-world code (including games) often includes at least some AVX instructions, so the AVX number is often more representative of "real" performance. The way AMD does it where it's a smooth transition based on current/thermals is definitely better than Intel's "whoops, AVX instruction, pump the brakes!".
But yes, if you can get higher clocks at least some of the time then you might as well.
The other other downside is that flipping between power states can cause problems/crashes too. It shouldn't but it can.
