Since the rise of SSDs, they have been using flash memory, also called NAND. Compared to traditional hard disks, this memory has much lower access times, where most of the speed gain comes from. In terms of sequential speed, the nano-chips are not particularly fast by themselves, but because they are controlled in parallel, high speeds can still be achieved.

The move from SATA SSDs to NVMe mainly brought up the speeds in workloads with a high queue-depth. In normal English: If there is a long queue of commands waiting, they can be processed very quickly thanks to the faster interface. What still went almost as fast were the assignments with shorter queues. The controller will not be able to perform smart tricks and is limited by the speed of the NAND.