Indeed, which is why I always compared wallclock time at equal power consumption. I think that's the fairest way to compare models.