2 ms·

5-6 per cycle per thread is more typical. Can be faster with a micro-op cache or a loop stream buffer.