4 ms·

Going by their estimates, no kidding - they're talking a 3x reduction in latency from their proposal.