Can You Hot-Swap NVMe Drives in a Server? What Hot-Plug Requires

Yes, NVMe drives can be hot-swapped, but hot-plug is a property of the whole platform, not just the SSD. Enterprise NVMe drives in U.2, U.3 and EDSFF form factors are designed to be inserted and removed under power. Whether your server lets you do that safely depends on its backplane, the PCIe ports behind it, the firmware settings, the operating system and sometimes a platform feature that manages the bays. The specifics depend on the platform, backplane and OS, so check the server's documentation before you pull a live drive. This guide explains what to look for and how to replace a drive without losing data.

Are NVMe SSDs hot-swappable?

It depends on the form factor first:

• U.2 and U.3 (2.5-inch, in a carrier): designed for hot-plug in a supporting backplane. Examples include the Samsung PM1733 1.92TB U.2 and the Kioxia CD8-R 1.92TB U.3, both PCIe 4.0 x4.
• EDSFF (E1.S, E3.S): designed from the start as hot-plug data center form factors.
• M.2: treat as not hot-swappable. M.2 slots on motherboards and ordinary adapter cards need the server powered off. A few dedicated boot modules place M.2 drives in hot-plug carriers, but that is a feature of the module, not of M.2.
• PCIe add-in card SSDs: treat as not hot-swappable unless the server's documentation specifically says otherwise.

Then it depends on the server. A drive built for hot-plug still needs a platform that supports it, and platforms differ in what they support: some handle only orderly removal, where you tell the system first; others also tolerate surprise removal, where the drive is pulled without warning. Assume orderly removal is required unless the documentation states otherwise.

What do I need for NVMe hot-plug in a server?

• A backplane built for NVMe hot-plug, either U.2 NVMe or U.3 tri-mode, with drive presence detection and status LEDs.
• PCIe ports (on the CPU or a PCIe switch) that support hot-plug, with the feature enabled in the BIOS or UEFI settings where the platform exposes it.
• Platform features some servers use to manage NVMe bays, such as Intel VMD on many Intel Xeon platforms, enabled and paired with the matching driver where required.
• An operating system or hypervisor version that supports NVMe hot-plug on that platform. Current Linux, Windows Server and VMware ESXi releases support it on qualified hardware; check the server vendor's OS support matrix.
• The server's management tool (iDRAC, iLO, XClarity or the vendor's equivalent) or OS commands to prepare a drive for removal and light its locate LED.
• If the NVMe drives sit behind a tri-mode RAID controller, the controller handles insertion and removal much as it does for SAS drives.

Where to confirm all of this: the server's technical guide or installation and service manual usually states NVMe hot-plug support for each backplane option, along with the removal procedure and the meaning of each bay LED. The OS vendor's hardware compatibility list confirms the operating system side. If the two disagree, or the manual only covers SAS and SATA bays, treat NVMe hot-plug as unsupported on that configuration and schedule a maintenance window for the swap instead. A short planned shutdown costs far less than a crashed host or a corrupted volume.

Is it safe to pull an NVMe drive while the server is running?

Safe when the platform supports it and you follow the removal sequence. Not safe when there is I/O in flight to the drive, when the drive holds the only copy of its data, or on a platform that does not support hot-plug, where a pulled drive can cause I/O errors, a hung device or a system crash. The safe sequence looks like this:

• Identify the drive by serial number and slot, and switch on its locate LED so the right carrier gets pulled.
• Stop I/O to it: remove it from its RAID set, ZFS pool or storage pool, migrate any virtual machines, and unmount its file systems.
• Prepare it for removal using the management tool or the OS. On Linux, if the platform tools do not do it, the device can be detached from the PCIe bus through sysfs before the drive is pulled.
• Wait for the bay's LED to show the drive is safe to remove. LED behavior varies by vendor, so check the server's guide.
• Pull the drive, insert the replacement, confirm the OS detects it, and start the rebuild or resilver.

Hot-plug is a platform feature, not a drive feature. The drive is built for it; the backplane, firmware and OS decide whether your server allows it.

Does hot-swap replace the need for RAID?

No. Hot-swap only means you can change a drive without powering down. If the drive you remove was the only copy of its data, the data leaves with it. Hot-swap delivers uptime when it is paired with redundancy: hardware or software RAID, a ZFS mirror or RAIDZ, erasure coding, or replication at the application layer. Our guides to RAID vs HBA and NVMe storage architecture cover how to build that redundancy on NVMe.

Can I add NVMe drives to empty bays while the server runs?

On platforms that support hot-plug, hot-add usually works too: the new drive appears to the OS as a new NVMe device, ready to be added to a pool or formatted. As with removal, confirm support in the server's documentation first, and check that the bay has the correct carrier and is cabled to a PCIe port. On some servers, bays are fitted with blanks but not wired for NVMe unless the NVMe backplane option was ordered.

What replacement NVMe drive should I order?

Match the form factor and the carrier first. Most U.3 drives also run in U.2 bays, so U.3 is a flexible choice; a U.2 drive generally will not run in a U.3 tri-mode bay. Then match or exceed the capacity of the drive being replaced, and keep endurance at the same class or higher. The PCIe generation adjusts itself: the Samsung PM1743 1.92TB is a PCIe 5.0 x4 U.3 drive rated up to 14,000 MB/s sequential read, but in a PCIe 4.0 bay it runs at PCIe 4.0 speed.

For older PCIe 3.0 servers, the Samsung PM983 960GB is a PCIe 3.0 x4 U.2 drive rated 1.3 DWPD. For capacity, the Micron 9400 PRO 7.68TB and Micron 6500 ION 30.72TB are PCIe 4.0 x4 U.3 read-intensive drives. The U.2 hot-plug and U.3 hot-swap pages list drives supplied with carriers.

Nexus Compute supplies new U.2 and U.3 NVMe SSDs through authorized distribution, with the full manufacturer warranty, and checks the form factor, carrier and PCIe generation against your server before dispatch. Send the server model and the failed drive's part number with a quote request, and we will reply within 48 business hours.