Apple put on the announce page that 4 Studios in this config can run inference at 3x the speed as 1 Studio.