3 ms·

Surely one per core is near optimal?