3 ms·

I guess you are talking about something different; I'm not really treating the cores as ML accelerators.