so i would take my performance figures somewhat as lower bounds - we know the actual memory macro is faster than this, but we don't know how much