Enterprise vs Consumer NVMe: What Actually Goes in Our Racks

Two NVMe drives with the same capacity and the same headline speed can behave completely differently under a production workload. The difference is not marketing: it is power-loss protection, sustained-write behavior, and endurance math. If you buy hosting without understanding this split, you are rolling dice with your database.

The three differences that matter

• Power-loss protection (PLP): enterprise drives carry onboard capacitors that flush cached writes to NAND when the power dies. Consumer drives simply lose the cache — and with it, the fsync guarantees your database assumed were real. This is the single biggest silent data-corruption risk in cheap infrastructure.
• Sustained vs burst writes: consumer drives post heroic peak numbers, then throttle once the SLC cache fills. Enterprise drives hold their steady-state write speed for hours, which is what a write-heavy workload actually experiences.
• Endurance (DWPD): drive endurance is rated in drive-writes-per-day. Consumer drives are rated around 0.3 DWPD; enterprise read-intensive drives start around 1 DWPD and mixed-use drives go to 3, 5, or beyond. Churn-heavy workloads burn through the difference fast.

What we deploy, and where

Our bare metal line is specced with enterprise-class NVMe as the default: the entry server carries 2 × 1 TB NVMe SSD, the 12-core/96 GB tier runs the same pair, and the storage-heavy configurations pair big NVMe with bulk SATA where capacity matters more than latency. The VPS line runs NVMe across the board — direct-attached for the local tier, Ceph-replicated NVMe when high availability matters more than a microsecond.

The principle is consistent: storage is the component whose failure mode is most expensive, so it is the last place to economize.

How to check what you are actually getting

• Ask for the exact drive model, not "NVMe SSD" — vendor part numbers tell you the tier.
• Look for PLP in the datasheet; if it is not listed, it is not there.
• Match DWPD to your write workload: daily writes ÷ drive capacity. If the number exceeds the rating, you are spending endurance you do not have.
• For databases and validators, demand sustained (not peak) write figures under a queue depth that resembles your workload.

The takeaway

NVMe everywhere is a genuine revolution — the throughput and latency floor it sets is why a $169 server today outperforms dedicated boxes that cost four figures a few years ago. But "NVMe" is a connector standard, not a quality tier. When a hosting price looks too good, the drives are usually the first place the savings were found.

We would rather show you the spec than explain the outage. Ask us what is in the machine you are buying — it is a fair question, and the answer should be boring.