4 ms·

I think it's mostly because while the internal parallelism of NVMe is fantastic our logical use of them is still largely sequential.