NVMe

SATA was designed for spinning hard drives — its speed limits reflect HDD performance expectations from the 1980s. As SSDs became capable of speeds far beyond what SATA could carry, NVMe was developed as a purpose-built interface for flash storage.

NVMe connects via PCIe lanes directly to the CPU, with a command queue designed for the parallelism of modern flash — 65,535 queues of 65,535 commands each, versus SATA's single queue of 32. The result: consumer NVMe drives reach 3-7GB/s sequential read; enterprise NVMe drives exceed 12GB/s. Random I/O latency drops below 100 microseconds.

Form factors

M.2 — a compact slot on the motherboard, used in laptops, workstations, and consumer hardware. Most new laptops use M.2 NVMe.

U.2 / U.3 — enterprise server form factor, physically similar to a 2.5" SATA drive but with PCIe signaling. Hot-swappable in enterprise server drive bays.

PCIe add-in card — full-length card in a PCIe slot. Used for maximum performance or when M.2/U.2 slots aren't available.

NVMe over Fabrics (NVMe-oF)

NVMe-oF extends NVMe over a network, allowing remote storage to be accessed with near-local latency. Used in high-performance storage networks as an alternative to iSCSI, with significantly lower latency. Increasingly relevant for AI infrastructure and high-performance computing.