Enterprise NVMe SSD — U.2, M.2, E1.S & PCIe Gen5
Complete buyer guide & DWPD reference

DWPD measures write endurance — how many times the full drive capacity can be overwritten per day over the warranty period (typically 5 years). A 3.84TB drive rated at 1 DWPD supports 3.84TB of writes per day for 5 years before reaching its rated endurance limit. Read-intensive (RI) drives are rated at 0.4 DWPD; mixed-use at 1–3 DWPD. Using a 0.4 DWPD drive in a write-heavy workload will exhaust the NAND cells prematurely and void warranty.

TLC (Triple Level Cell) stores 3 bits per NAND cell — more durable and faster writes, but higher cost per TB. QLC (Quad Level Cell) stores 4 bits per cell — 25–33% cheaper per TB, but lower write endurance (typically 0.1–0.3 DWPD) and slower sustained writes. Enterprise QLC drives (used in read-intensive arrays like Pure Storage FlashArray//C) use large DRAM caches and SLC write buffers to mask the QLC write penalty in read-dominated workloads.

Consumer NVMe drives (Samsung 990 Pro, WD Black) lack power loss protection (PLP), are rated for ~0.3 DWPD, have no hot-swap support, and have MTBF of ~1.5M hours. Enterprise NVMe drives (Samsung PM9A3, Micron 9400) include PLP capacitors that flush write buffers during power failure, are rated for 1–3 DWPD, support hot-swap (U.2), and carry 2M+ hour MTBF with 5-year warranties. Never use consumer SSDs in production servers.

PCIe Gen4 x4 provides 8 GB/s of raw bandwidth — enough for sequential reads up to ~7 GB/s (Samsung PM9A3). PCIe Gen5 x4 provides 16 GB/s — unlocking drives like the Kioxia CM7-R at 13.9 GB/s. PCIe Gen5 requires a Gen5-capable server (Intel Xeon 6, AMD EPYC 9005) and backplane. Gen5 drives will physically fit in a Gen4 backplane but will run at Gen4 speeds — you pay for Gen5 performance and receive Gen4. Confirm your backplane generation before ordering.

U.2 (SFF-8639, 2.5" 15mm) is the universal enterprise form factor — hot-swappable, supports drives up to 25.6TB, fits all mainstream server drive bays. M.2 is smaller, not hot-swappable, and limited to lower-capacity drives; used for boot/OS drives and compact edge servers. EDSFF (E1.S, E3.S — "Ruler" format) is the emerging high-density format in newer server models — higher drive count per U, more efficient thermal management, but requires a compatible EDSFF backplane.

NVMe-oF extends the NVMe protocol over a network fabric (RDMA/RoCE, Fibre Channel, TCP) to enable sub-microsecond access to remote NVMe storage. It eliminates the SCSI translation overhead of traditional iSCSI or FC-SCSI SAN, bringing NVMe's latency advantage to networked storage. Requires NVMe-oF capable HBAs or NICs on the server side and NVMe-oF capable storage arrays (Pure Storage FlashArray, NetApp AFF, etc.) on the target side.