3 ms·

I think in this case it has more to do with the multiple-CCX design and the synchronization bottleneck it creates than single-threaded performance.