Storage · Explainer
NVMe vs SSD vs HDD in a server, by the numbers.
NVMe vs SSD vs HDD is two questions: what holds the data, and how the drive talks to the processor. A hard drive keeps it on a spinning platter and moves a head to read it. A SATA SSD keeps it in flash but stays on the old disk interface. An NVMe SSD puts the same flash on PCIe lanes and speaks a protocol written for it, so it answers in microseconds and runs thousands of requests at once.
- Figures from the specs
- Server workloads
- Checked October 2026
Below: what NVMe actually is, the bus ceilings of each interface, why random reads separate the three far more than sequential speed does, how each type fails, and which drive a database, a VM host or a backup box needs.
Definitions
What NVMe storage is, and what it is not
NVMe (NVM Express) is a protocol, not a type of memory. An NVMe drive is an SSD: the same NAND flash you find in a SATA SSD, attached to PCIe lanes instead of a SATA port, and driven by a command set designed for flash rather than for spinning disks.
So "NVMe vs SSD" is a slightly wrong question. The real comparison is NVMe vs SATA SSD: same medium, different road to the processor. The road is what sets the ceiling. SATA and its driver model, AHCI, were built in the years of hard drives, when a disk could only serve one request at a time anyway.
SAS sits in between. It is the enterprise disk interface, twice the line rate of SATA per lane and dual-ported, and it carries both hard drives and SSDs. Between them, our server lines take NVMe, SATA and SAS drives; the catalogue section below shows which machine takes what.
- MediumSpinning platterNAND flash
- InterfaceSATASASPCIe
- ProtocolAHCISCSINVMe
- Form factor3.5 in2.5 inM.2U.2E1.S
Bandwidth
NVMe vs SATA SSD: the bus sets the ceiling
A SATA SSD cannot read faster than about 600 MB/s, because SATA III runs at 6 Gb/s and loses a fifth of that to encoding. An NVMe drive on four PCIe 3.0 lanes has about 3.9 GB/s to work with, which is where the familiar "six times faster" figure comes from. PCIe 4.0 doubles that, and PCIe 5.0 doubles it again.
Interface ceilings from the SATA, SAS and PCI Express specifications after line encoding (8b/10b for SATA and SAS-3, 128b/130b for PCIe). The hard drive figure is a current 7,200 rpm enterprise datasheet, outer tracks. A real drive reaches somewhat less than its bus allows.
Sequential speed matters when you copy large files: restoring a backup, moving VM images, reading video. For most server work it is not the number that hurts. That is the next section.
Latency and IOPS
Random reads are where the three drives part ways
A database, a mail server or a set of virtual machines reads and writes small blocks scattered all over the disk. Each request on a hard drive waits for the head to move and the platter to turn under it; on flash it waits for nothing mechanical. That is why the gap in random reads is thousands of times, not six.
Half a turn of a 7,200 rpm platter, the average wait before the right sector arrives. Add the head's seek and a hard drive manages a couple of hundred random reads a second.
AHCI, the SATA driver model: one command queue, 32 commands deep. Every processor core has to line up for it.
NVMe: up to 65,535 I/O queues, each up to 65,536 commands deep. Linux usually gives each core its own queue, so the cores do not wait on each other.
Datasheets tell the same story. A current 7,200 rpm enterprise hard drive lists about 170 random 4K reads a second. An enterprise SATA SSD lists close to 100,000. An enterprise PCIe 4.0 NVMe drive lists around a million. Your own numbers will be lower, and depend on block size and queue depth, so measure with fio on the machine you rent rather than trusting anyone's chart, ours included.
Reliability
How each drive wears out
Hard drives fail mechanically: bearings, heads, a motor. Flash fails by use: every cell survives a limited number of writes. Neither is "more reliable" in general; they fail differently, and you plan for each in a different way.
Hard drives: watch the mechanics
A spinning drive does not care how much you write, but it does care about heat, vibration and age. SMART counters such as reallocated and pending sectors usually move before it dies. Keep them in your monitoring.
SSDs: watch the writes
Flash endurance is printed on the datasheet as TBW (terabytes written) or DWPD (drive writes per day over the warranty). Read the wear level with
smartctl -aornvme smart-log. A read-heavy web server will never reach it; a busy database log might.Power-loss protection is a drive feature
Capacitors that flush the drive's cache when power drops come with enterprise SSDs, SATA or NVMe alike. They are not part of the NVMe protocol, and many consumer NVMe drives lack them. Check the datasheet of the drive you get if your database relies on synchronous writes.
RAID and backups, whatever the drive
On Linux, NVMe drives are usually mirrored in software with
mdadmor ZFS rather than behind a RAID card. A mirror keeps the server up when one drive dies. It is not a backup: a deleted table is deleted on both drives.
Sizing
NVMe for servers: which workloads need it
Pick NVMe when the workload does many small random reads and writes, or when several cores hit the disk at once. Pick a SATA SSD when the data mostly sits in memory after the first read. Pick hard drives when you need many terabytes and read them in long runs.
| Workload | Drive | Why |
|---|---|---|
| Databases: PostgreSQL, MySQL, MongoDB | NVMe | Small random reads, a write-ahead log that waits on every commit, and many connections at once. |
| Virtual machine host, Proxmox or KVM | NVMe | Every guest's disk is random I/O to the host, and the guests queue up together. |
| Build servers, CI runners | NVMe | Thousands of small files written, read and deleted on every build. |
| Web and application servers | SATA SSD or NVMe | Code and hot data live in RAM after the first request; the disk is rarely the limit. |
| Media libraries, video on demand | SATA SSD or hard drives | Large files read front to back. The network port is usually the bottleneck before the disk is. |
| Backups and archives | Hard drives | Long sequential writes, rare reads, and the lowest cost per terabyte. |
Larger servers often mix them: the operating system and the database on flash, bulk data on hard drives in the same chassis. Hard drives still cost the least per terabyte, and that is the main reason to keep them. For database machines see SQL dedicated servers, and for VM hosts Proxmox servers.
Live from the catalogue
The drives our dedicated servers take
3 of the 18 server lines in our configurator take NVMe SSDs, up to 3.84 TB a drive. These figures come straight out of the configurator's catalogue and move whenever it does.
| Drive | Interface | Server lines | Capacity per drive |
|---|---|---|---|
| NVMe SSD | PCIe, NVMe protocol | 3 | 238 GB to 3.84 TB |
| SAS SSD | SAS | 9 | 15.3 TB to 30.72 TB |
| SATA SSD | SATA, 6 Gb/s | 15 | 32 GB to 15.36 TB |
| SAS hard drive | SAS, spinning disk | 1 | 2 TB |
| SATA hard drive | SATA, spinning disk | 7 | 500 GB to 2 TB |
Machines that take NVMe
Each one opens its configurator, where you choose the drives, their number and the array.
Answers
NVMe vs SSD vs HDD: common questions
Is NVMe worth it for a server?
For databases, virtual machine hosts and build servers, yes: they live on random I/O, and NVMe serves it thousands of times faster than a hard drive and several times faster than a SATA SSD. For a web server that keeps its working set in memory, a SATA SSD performs almost the same. Check that the machine takes NVMe before you plan around it: not every chassis has the slots.
SATA or NVMe for a server: which should I choose?
Choose NVMe if the server runs a database, hosts virtual machines or handles many users writing at once. Choose SATA SSD if the machine only allows SATA drives or the workload is light and read-mostly. Choose SATA hard drives for capacity, such as backups and archives.
Is NVMe faster than an SSD?
An NVMe drive is an SSD. What is faster is the interface: an NVMe SSD on four PCIe 3.0 lanes has about 3.9 GB/s of bandwidth against about 600 MB/s for a SATA SSD, and far deeper command queues. On PCIe 4.0 and 5.0 the bandwidth doubles each generation.
What is an NVMe storage server?
A server whose drives are NVMe SSDs connected over PCIe, usually in U.2, E1.S or M.2 form. It can be a single machine running a database on local NVMe, or a box full of NVMe drives that shares its storage with other servers over the network.
What is NVMe SAN, or NVMe over Fabrics?
NVMe over Fabrics (NVMe-oF) carries the NVMe command set across a network instead of a PCIe slot, over RDMA, Fibre Channel or plain TCP. A SAN built on it lets many servers use shared flash with less overhead than iSCSI. Our dedicated servers come with their drives inside the chassis, which is the lowest-latency option because no network sits between the processor and the flash.
How much faster is NVMe than a hard drive?
For sequential reads, roughly 14 to 55 times: a 7,200 rpm drive reads under 300 MB/s while NVMe drives on PCIe 3.0 to 5.0 reach about 3.9 to 15.8 GB/s. For random 4K reads the gap is far bigger: about 170 a second from a hard drive against around a million on an enterprise PCIe 4.0 NVMe datasheet.
Can one server have both NVMe and hard drives?
Yes, when the chassis has bays for both and the controller supports them. A common layout puts the operating system and the database on NVMe and bulk data on hard drives. Not every chassis can: small cartridge servers often have room for one or two drives of a single kind. The configurator for each of our machines lists the drive families it accepts.
Elsewhere on the site
Where to go from here
Put the database on NVMe
Pick the machine, then the drives. Unsure about the drive? Put the question to the chat. The AI replies within seconds, looks up your account and machines if you want, and hands you to an engineer on request.