CMR vs SMR Drives for NAS: Why SMR Wrecks RAID Rebuilds

Buy the wrong hard drive for a NAS and everything looks fine until the day a disk dies. Then you swap in a replacement, the array starts to rebuild, and the rebuild crawls for days, stalls, and sometimes throws the new drive out entirely. The culprit is almost always one hidden spec the box never advertised: the recording technology. A drive-managed SMR disk in a RAID or ZFS array is a slow-motion data-loss trap, and most buyers never know until they are mid-rebuild.

This guide breaks down CMR vs SMR for a NAS: what each recording type actually is, why SMR specifically breaks array rebuilds, how to tell the two apart before you pay, and which drives are safe to buy. NAS drive prices are climbing hard in 2026 on the back of the AI storage shortage, so picking the right drive the first time matters more than usual.

Verified against each manufacturer’s current CMR/SMR drive list in June 2026.

The short answer

For any NAS that runs RAID, ZFS, Synology SHR, Unraid parity, or any redundant array: buy CMR drives, and avoid drive-managed SMR. The NAS-branded lines are CMR across the board, so the safe move is to stay inside them:

• Safe (CMR): Seagate IronWolf and IronWolf Pro, WD Red Plus and WD Red Pro, Toshiba N300 and N300 Pro.
• Avoid in an array (drive-managed SMR): plain WD Red with the EFAX code (2 to 6TB), WD Blue 3.5-inch (EZAZ), Seagate BarraCuda 4TB and 8TB (3.5-inch), Toshiba P300 4TB and 6TB.

That single rule prevents the multi-day rebuild horror story. The rest of this guide is why it is true and how to verify any drive yourself.

What CMR and SMR actually are

Both are ways of laying magnetic tracks on a spinning platter, and the difference is how close the tracks sit.

CMR (conventional magnetic recording) writes each track in its own lane with a small guard band between tracks. The write head is wider than strictly necessary for reading, so the lanes cannot touch. Every track can be rewritten in place without disturbing its neighbours. This is the traditional layout, and it is what every hard drive used before density pressure forced a change.

SMR (shingled magnetic recording) overlaps the tracks like roof shingles to pack more data onto the same platter. The trick works for reading, but it breaks in-place writes: because each track partially covers the one beside it, writing to a track also clobbers the adjacent track. To change data in the middle of a shingled zone, the drive has to read the surrounding tracks, modify them, and rewrite the whole sequence. A single small write can trigger a cascade of rewrites across a band of tracks.

There are two flavours of SMR, and the distinction is the whole story:

• Host-managed SMR (HM-SMR) exposes the shingled structure to the operating system, which then writes strictly sequentially and never randomly. This is used in hyperscale cold storage (Seagate’s 32TB-and-up Exos host-managed drives, for example) where the software is built for it. It is not what ends up in a home NAS.
• Drive-managed SMR (DM-SMR) hides the shingling behind the firmware so the drive looks like any ordinary disk to the host. The OS often cannot even detect that it is SMR. This is the kind that lands in consumer drives, and the kind that wrecks rebuilds, precisely because nothing warns you it is there.

Why SMR wrecks a RAID or ZFS rebuild

A DM-SMR drive copes with everyday use by absorbing writes into a small CMR cache zone, then quietly folding them into the shingled tracks during idle time. For a desktop that writes in short bursts, that works. An array rebuild is the exact opposite workload.

When a disk fails and you insert a replacement, the array reconstructs every block on the new drive. That is sustained, write-heavy I/O spread across the entire disk with no idle gaps. On a DM-SMR drive the CMR cache fills within minutes, and from then on the firmware is locked into continuous read-modify-write of shingled zones. Throughput collapses from the usual 150-plus MB/s down to single digits or low tens of MB/s, and a rebuild that should take hours stretches into days. ServeTheHome’s RAIDZ testing found an SMR drive left the array exposed 13 to 16 times longer than a CMR drive, with a resilver running close to nine days.

Western Digital itself put it plainly when it explained the difference: “sustained random writes during ZFS resilvering causes a lack of idle time for DMSMR drives, resulting in significantly lower performance.”

The slow rebuild is not even the worst case. When the firmware pauses host I/O to reorganise its shingles, that pause can outlast the timeout the controller or kernel allows. The system then concludes the drive is dead. On ZFS this is documented: iXsystems found the 4TB WD40EFAX “returns IDNF errors, becomes unusable, and is treated as a drive failure by ZFS” under resilver load. So the new drive you added to heal the array gets ejected from it instead, and a single-disk failure can spiral into a second one while the data is unprotected.

This is where the NAS-drive feature TLER (Time-Limited Error Recovery), called ERC on Seagate and Toshiba, earns its keep. NAS-class CMR drives cap their internal error recovery at a few seconds so the controller never times out and ejects them. Desktop SMR drives lack that short limit, their internal shingle management stalls run long, and RAID marks them unreliable. On Linux software RAID the community mitigation is to raise the block-layer timeout so the kernel waits the stall out:

echo 180 | sudo tee /sys/block/sdX/device/timeout

That buys time against the default 30-second timeout, but it is a workaround, not a fix. The drive is still glacially slow during the rebuild and still the wrong tool for the job. The real fix is to not put DM-SMR in the array.

The 2020 WD Red scandal that started the panic

For years, knowing a drive’s recording type was a non-issue because NAS drives were all CMR. That changed in 2020, when buyers discovered Western Digital had quietly switched several WD Red models, the 2, 3, 4, and 6TB drives carrying the EFAX code, to drive-managed SMR without changing the name, the marketing, or the price. People who bought them as straight CMR replacements hit exactly the failed-resilver behaviour above.

The backlash forced two changes. Western Digital admitted in its own statement on WD Red drives that “we typically specify the designed-for use cases and performance parameters and don’t always talk about what’s under the hood,” and it split the line into three clear tiers that survive today:

• WD Red (plain, the budget line): DM-SMR, sold for light SOHO use, not for ZFS or write-heavy arrays.
• WD Red Plus: CMR, the drive WD explicitly points at ZFS, sold from 2 up to 14TB.
• WD Red Pro: CMR, higher RPM and workload rating, up to 26TB.

A US class action followed and settled for $2.7 million, covering buyers of the WD20EFAX, WD30EFAX, WD40EFAX, and WD60EFAX drives, with Western Digital also agreeing to label SMR on packaging going forward. The lasting lesson is simpler than the legal fallout: the recording type is a spec worth checking yourself, because it has been hidden before.

How to tell CMR from SMR before you buy

Three checks, fastest first:

1. Buy a NAS-branded line and you are already safe. IronWolf, IronWolf Pro, WD Red Plus, WD Red Pro, N300, and N300 Pro are CMR across every capacity. The only NAS-adjacent trap is the plain WD Red, so reach for Red Plus or Red Pro, never bare “Red”.
2. Check the manufacturer’s published list. Seagate maintains an official CMR and SMR drive list, Toshiba published one, and WD documents it on its support site. If a drive is not clearly on a CMR list, treat it as suspect.
3. Decode the model number. On WD, the EFAX code means DM-SMR, while EFRX, EFZX, EFPX, EFGX, EFBX, KFGX, and FFBX are CMR. WD Blue desktop drives ending EZAZ are DM-SMR. The exact code is printed on the drive label and in the spec sheet.

The 2026 CMR vs SMR cheat sheet

The recording type for the drives a NAS buyer is likely to weigh, current as of June 2026:

If you just want the short list of safe drives to buy, three CMR disks cover almost every NAS. The Seagate IronWolf Pro at 12TB is the value sweet spot for a home array, the WD Red Plus suits smaller two to four bay builds, and the IronWolf Pro 24TB packs the most capacity per bay when slots are scarce. All three are CMR, NAS-rated, and confirmed against the manufacturer lists above. NAS drive prices are climbing week to week in 2026, so check the live price before you commit.

For the full comparison of capacities, workload ratings, warranties, and current pricing across every CMR family, see the companion best NAS hard drives guide. If you have not settled on the box itself yet, the best NAS for home and self-hosting roundup pairs each enclosure with the drives that fit it. If you are building the array on ZFS, the ZFS RAID levels guide covers how mirror and RAIDZ choices change rebuild time and risk, which is the other half of this decision.

SMR is not evil, it is just the wrong job

None of this means SMR drives are bad. Shingled recording is genuinely good for data that is written once and rarely changed but needs to be read efficiently from anywhere: append-only logs, write-once-read-many archives, and cold backups. Dropbox built its append-only storage on SMR, and host-managed SMR is a backbone of hyperscale cold tiers. The problem is never the technology, it is putting a drive-managed SMR disk into a random-write, no-idle-time role like an array member.

So if you already own SMR drives, do not bin them. They make perfectly good single-drive backup targets, an offline copy you write to occasionally, or bulk media storage that you fill once and mostly read. Just keep them out of the RAID, ZFS, or SHR pool that has to rebuild under load.

If you already have SMR drives in an array

If a rebuild is crawling and you suspect SMR, confirm the model against the lists above, then plan a one-at-a-time migration to CMR: replace a suspect drive, let the array fully resync on the new CMR disk (budget extra time, because the remaining SMR members are still slow), and repeat until the pool is all CMR. On Linux software RAID the timeout tweak above can keep the kernel from ejecting a stalling SMR drive long enough to finish a migration, but treat that as triage. The destination state is an all-CMR array, whether you run it on a turnkey box, on TrueNAS, or as a Proxmox ZFS pool.

The decision really is that simple. For anything that rebuilds, buy CMR, stay inside the NAS-branded lines, and verify the model code before you check out. Spend ten seconds on the spec the box hides and you will never meet the nine-day resilver.