CMR vs SMR NAS Drives: What Actually Changes

Two drives can share a capacity, an interface, a rotational speed and a price bracket and still behave completely differently inside a parity array. The difference is how the data is laid down on the platter, and it is the single specification most worth checking before a NAS drive purchase.

What the two recording methods do

Conventional magnetic recording, usually written CMR, places each track next to the previous one with a gap between them. Every track can be rewritten independently, because writing to one does not disturb its neighbours.

Shingled magnetic recording, SMR, removes that gap. The write head lays down a wider track than the read head needs, so tracks are deliberately overlapped like roof shingles, each one partially covering the one before it. That overlap is where the extra areal density comes from, and Western Digital’s own zoned storage documentation describes the arrangement in exactly those terms.

The cost is asymmetric. Reading a shingled track is unaffected. Rewriting one is not, because the track physically overlaps its neighbours, so modifying it means rewriting the whole band it belongs to. Drives group tracks into zones or bands to bound how much has to be rewritten, and drive-managed SMR products hide the whole mechanism behind a persistent cache region that absorbs incoming writes and reorganises them into shingled bands later, during idle time.

That cache is the crux of the entire argument. While it has room, a drive-managed SMR disk behaves like any other drive. When it fills and the drive has no idle time in which to drain it, throughput falls to whatever the reorganisation process can sustain, and it stays there until the workload stops.

Where it goes wrong in a NAS

The specific workload that empties an SMR write cache and never lets it refill is a parity rebuild.

When a drive in a parity array fails and is replaced, the array reconstructs the missing data onto the new drive by reading everything from the surviving members. On a ZFS pool this is a resilver, which OpenZFS documents as the operation zpool replace starts on the incoming device; on a conventional RAID set it is a rebuild. Either way the replacement drive receives sustained writes for as long as the operation lasts, which for large modern drives is measured in many hours and often days, and the operation offers the receiving drive no idle time at all.

A drive-managed SMR disk in that position exhausts its cache early and then completes the rest of the rebuild at its shingled-rewrite rate. Rebuilds that would take a day stretch out much further. That matters for more than patience: an array running on reduced redundancy is exposed for the whole duration, and stretching the exposure window is precisely the risk that parity exists to shorten.

The same mechanism shows up in three other places, less dramatically. Long sustained writes, such as the initial fill of a new array or a large restore, run fast until the cache is gone and then drop. Random rewrite workloads, such as a busy virtual machine image or a database file, never let the drive catch up. And snapshot-heavy or checksum-scrubbing filesystems generate background rewriting the drive would rather not be doing.

How this became a labelling problem

Until 2020 the recording technology of consumer drives was frequently absent from published specifications. In April of that year Western Digital published a statement acknowledging that its lower-capacity WD Red drives, the 2 TB to 6 TB models, used drive-managed SMR, while 8 TB and above and everything at 1 TB or below remained CMR. The industry moved to stating recording technology on datasheets after that, and both major NAS vendors now publish it per model.

The practical result is a split product line, and both vendors now make it checkable. Western Digital separated its NAS drives into WD Red Plus and WD Red Pro: the March 2025 WD Red Plus product brief lists every model from the 2 TB WD20EFPX to the 12 TB WD120EFGX as CMR, with no exceptions in the table. Seagate publishes a per-family CMR and SMR model list, and on it IronWolf shows CMR at 2, 4, 6, 8, 10, 12 and 16 TB with no SMR models at all, and IronWolf Pro shows CMR from 2 TB to 24 TB with no SMR models either. Seagate’s shingled drives sit elsewhere in the range: BarraCuda at 4 TB and 8 TB, Archive at 8 TB, and the highest-capacity Mozaic Exos models.

None of that makes SMR a defective technology. It makes it a technology with a workload profile, sold for years without stating the profile. The lesson is not that one brand is safe and another is not; it is that the answer is now published per model number, so there is no longer any reason to guess at it.

The comparison in one table

	CMR	SMR (drive-managed)
Track layout	Adjacent tracks with a guard gap	Overlapping tracks in bands
Areal density	Lower for a given platter count	Higher, which is the entire point
Sequential read	Full speed	Full speed
Sequential write	Full speed, sustained	Full speed until the cache fills, then much slower
Random rewrite	Full speed	Requires rewriting a whole band
Parity rebuild or resilver	Predictable, bounded by drive throughput	Cache exhausts early, rebuild time can multiply
Good fit	Any parity array, virtual machine storage, general NAS use	Write-once archive, external backup, sequential cold storage
How to identify	Stated on current datasheets	Stated on current datasheets; older stock may not say

How to check a specific drive

Model number first. Both major vendors publish recording technology per model number in the product brief or datasheet, not in the retail listing, and the retail listing is frequently wrong or silent. Find the exact model string, including the suffix, and check it against the manufacturer’s own document rather than a marketplace bullet list.

Seagate makes this easy: its CMR and SMR hard drive list gives the recording technology by family and capacity on one page. Western Digital puts it in the per-family product brief, so start from the WD Red Plus page and open the linked brief.

Suffix changes matter more than most people expect. Within a single capacity, a different suffix can mean a different platter count, a different rotational speed and, historically, a different recording method. The 12 TB WD Red Plus is the clearest example: WD120EFGX and WD120EFBX are both 12 TB, both CMR and both 7200 RPM, but the EFGX is air-filled with a 512 MB cache and an internal transfer rate quoted up to 260 MB/s, while the EFBX is helium-filled with 256 MB and up to 196 MB/s. Their power figures differ too. Two drives, one capacity, one product name, measurably different machines.

Two further checks are worth the minute they take. Drives bought secondhand or from grey-market sellers may predate the labelling changes entirely, so verify the model rather than the family. And a NAS vendor’s compatibility list confirms that a drive is supported by that appliance, which is a different question from whether it is CMR.

What to do with SMR drives you already own

They are perfectly good for the workload they were designed for. A shingled drive is a reasonable external backup target, a reasonable destination for a write-once archive, and a reasonable place to park media that is written once and read many times. What it should not be is a member of a parity array that will one day have to rebuild.

If an SMR drive is already inside an array, the sensible response is not panic but planning: replace it at the next opportunity, and make sure the array is not relying on it as the replacement target during a future rebuild. Repeated resilver failures on shingled members are the failure mode that turns a single drive fault into a lost pool.

Where this sits in a build

Recording technology is one of a small number of drive specifications that genuinely change how a machine behaves. The others are the annualised workload rating and the non-recoverable read error rate, both of which are on the same datasheets. Together they are the difference between a NAS drive and a desktop drive of the same capacity.

For how many of these drives to buy and in what arrangement, see the NAS buying guide. For how bay count, controller lanes and power interact once the drive count is settled, see sizing NAS hardware. And if an existing array has become slow rather than being born slow, an exhausted SMR cache is one of several candidates worth ruling out in NAS slow transfer speeds.

The rule is simple enough to apply without a spreadsheet. If the drive will ever be rebuilt into, buy CMR.