benchmarks. the perception of performance is almost entirely related to latency at low queue depths.

You cant tell the difference between 100 us and 1 ms.