CMR vs SMR Drives: Why It Decides NAS Reliability
Use CMR drives in a NAS. Never put SMR drives in a RAID array. CMR (conventional magnetic recording) writes each track independently; SMR (shingled) overlaps tracks to gain density, so rewrites force the drive to read-modify-write whole zones. Under the sustained writes of a RAID rebuild, SMR drives slow so drastically they can be dropped from the array as "failed" — the exact failure mode behind the 2020 WD Red controversy. The decoder rule: WD Red Plus/Pro and Seagate IronWolf are CMR; plain WD Red 2–6TB is SMR. For anything else, check the manufacturer's published CMR/SMR list before buying.
How do CMR and SMR recording actually work?
A hard drive stores data on concentric magnetic tracks. In CMR (conventional magnetic recording, also called PMR), each track sits separately with its own space: the drive can rewrite any track, any time, without touching its neighbors. Write performance is predictable and stays close to the rated figure under sustained load.
SMR (shingled magnetic recording) overlaps tracks like roof shingles. Because a drive's write head is wider than its read head, tracks can be packed partially on top of each other and still be read back — which squeezes meaningfully more capacity from the same platters at lower cost per terabyte. The catch: writing to one track disturbs the shingled track next to it. The drive can't rewrite a single track in place; it must read an entire zone of overlapping tracks, modify it in memory, and write the whole zone back. To hide this, SMR drives stage incoming writes in a small conventional (CMR-style) cache region and reshuffle zones during idle time.
Why does SMR exist, and where is it fine?
SMR is not a scam — it's a legitimate density/cost tradeoff, and for the right workload it works well. Reads perform essentially like CMR. Workloads that write once and read many times never hit the read-modify-write penalty. SMR is a reasonable choice for:
- Archival and cold storage — data written once and rarely modified.
- USB backup drives — periodic backup dumps with long idle periods for the drive to tidy zones.
- Secondary copies in a 3-2-1 backup scheme, where write speed isn't critical.
The problem is never that SMR drives exist. It's when SMR ends up — especially unlabeled — in workloads built on sustained rewrites. Like a NAS.
Why is SMR wrong for a NAS or RAID array?
Two compounding failure modes:
1. Sustained-write collapse. An SMR drive looks fast while its conventional cache zone absorbs writes. Fill that cache with a long continuous write — a big media transfer, an initial backup, an array resync — and throughput drops to a small fraction of the rated speed while the drive performs read-modify-write cycles on shingled zones in the background. The drive isn't broken; it's doing exactly what shingling requires. But your "NAS-class" transfer crawls.
2. The RAID rebuild disaster. A rebuild is the worst possible SMR workload: when you replace a failed drive, the array must write the reconstructed contents of an entire drive — many hours to days of continuous sequential writes — onto the new disk, with no idle time to reorganize zones. SMR drives have slowed so drastically under this load, and responded so sluggishly, that RAID controllers and software arrays have timed them out and dropped them as "failed" — a healthy drive ejected from the array mid-rebuild. That turns a routine drive swap into a degraded array running with no redundancy, during the exact window when a second drive failure destroys everything. This is why the rule is absolute: no SMR in a RAID array.
What was the 2020 WD Red controversy?
In 2020, the storage community discovered that Western Digital had quietly switched some capacities of the WD Red — a drive line explicitly marketed for NAS use — to SMR, without labeling the change. Users saw exactly the failure modes above: RAID rebuilds crawling or failing, and drives dropping out of arrays. The backlash was industry-wide, and it forced two lasting changes: WD split the line into plain WD Red (SMR) and WD Red Plus / Red Pro (CMR), and the major manufacturers — WD, Seagate, and Toshiba — began publishing which of their drives use CMR versus SMR. That disclosure is the durable win: today you can, and should, verify recording technology before buying.
CMR vs SMR at a glance
| CMR (conventional) | SMR (shingled) | |
|---|---|---|
| Track layout | Separate, independent tracks | Overlapping tracks, grouped in zones |
| Rewriting data | Any track rewritten in place | Whole zone read-modify-write |
| Sustained write speed | Predictable, near rated speed | Collapses once the cache zone fills |
| Read speed | Normal | Comparable to CMR |
| RAID rebuild behavior | Steady; rebuild completes in the expected window | Can slow drastically; drives have been dropped from arrays as "failed" |
| Why it exists | The default recording method | More capacity per platter, lower cost per TB |
| Good for | NAS, RAID, any rewrite-heavy workload | Archival, write-once data, USB backup drives |
| Verdict for NAS | Required | Never in a RAID array |
Which drives are CMR? The decoder rule
The well-documented cases you can rely on:
- WD Red Plus and WD Red Pro — CMR. This is what the "Plus" exists to signal.
- Plain WD Red, 2–6TB models — SMR. Do not put these in an array.
- Seagate IronWolf and IronWolf Pro — CMR. Seagate's NAS line has stayed conventional.
Beyond those, hedge: model lines change by capacity and revision, and desktop or external drives are frequently SMR without prominent labeling (shucked external drives are a classic trap). The universal advice — the thing that outlives any decoder table — is to check the manufacturer's published CMR/SMR list against your exact model number before buying. WD, Seagate, and Toshiba all publish one now.
While you're verifying: a NAS-rated drive buys you more than CMR. WD Red Plus and Seagate IronWolf are rated for 24×7 operation and carry firmware tuned for multi-drive vibration (NASware / AgileArray) — conditions a desktop drive isn't specified for.
What does the failure data say?
For drive reliability generally, the standard large-scale public source is Backblaze's published drive-stats reports, which track annualized failure rates across a fleet of well over two hundred thousand drives. The pattern: failure rates are low but never zero — commonly in the low single-digit percentages per year, varying by model and age — with failures clustering early in life and again in old age. The SMR lesson sits on top of that: since drives will fail, your array will eventually run a rebuild, and a rebuild that takes days instead of hours (or ejects the new drive) is exactly the widened second-failure window you built RAID to avoid. And whatever the drives — RAID is not a backup; keep a tested 3-2-1 backup.
Related reading
Frequently Asked Questions
CMR (conventional magnetic recording) writes each data track separately, so any track can be rewritten without touching its neighbors. SMR (shingled magnetic recording) overlaps tracks like roof shingles to pack in more capacity — but because writing one track disturbs the next, rewriting existing data forces the drive to read, modify, and rewrite an entire zone of tracks. Reads perform similarly on both; sustained and random writes are where SMR falls apart.
Two reasons. First, sustained writes collapse: SMR drives absorb bursts into a small conventional cache zone, and once it fills, write speed drops to a fraction of the rated figure while the drive reshuffles zones in the background. Second, RAID rebuilds are exactly that worst case — many hours to days of continuous sequential writes to one drive. SMR drives have slowed so drastically under rebuild load that arrays have timed them out and marked them as failed, turning a routine drive swap into a degraded-array emergency.
Yes — SMR is not a defective technology, it is a density/cost tradeoff. For write-once, read-mostly workloads it is fine: archival storage, a USB backup drive that receives periodic dumps, cold media libraries, or a secondary copy in a 3-2-1 backup scheme. The problem is specifically sustained rewrites and RAID rebuilds, which is why SMR does not belong in a NAS array.
In 2020 it emerged that Western Digital had quietly used SMR in some WD Red drives — a line explicitly marketed for NAS use — without labeling it. Users hit failed and crawling RAID rebuilds, the story blew up across the storage community, and after the backlash WD split the line: plain WD Red (2–6TB models) stayed SMR, while WD Red Plus and Red Pro were designated CMR. It remains the standard cautionary tale for why you check recording technology before buying NAS drives.
The well-documented cases: WD Red Plus and WD Red Pro are CMR; plain WD Red in the 2–6TB range is SMR; Seagate's IronWolf and IronWolf Pro NAS line is CMR. Beyond those, don't guess from the product name — capacities and sub-models change. Every major manufacturer (WD, Seagate, Toshiba) now publishes a CMR/SMR list on its website after the 2020 controversy; check the exact model number against that list before buying.
The core problem is behavior under load, not raw failure rate — an SMR drive can be perfectly healthy and still get dropped from an array because it responds too slowly during a rebuild. For actual failure-rate data, Backblaze's published drive-reliability reports are the standard large-scale source: annualized failure rates are typically in the low single-digit percentages, varying by model and age. The danger SMR adds is stretching a rebuild from hours to days, widening the window in which a second drive failure can kill the array.