3 ms·

as i understand it's compute, just not cpu compute, those cpu are designed to be good enough for cuda servers