Shingled Magnetic Recording has a terrible reputation, and most of it is deserved — in the wrong workload. The perceived wisdom goes like this: SMR drives write quickly into a small CMR-managed cache region, and once that fills, performance falls off a cliff as the drive starts reshingling in the background. Transfer speeds collapse, progress bars stall, and you regret your purchase.
I wanted to know what that actually looks like in practice, so I ran an experiment: a single sustained transfer of 533 GB of render output — large sequential files, roughly 200 MB per frame, spread across several folders — onto a Western Digital My Passport 6TB, a known SMR drive. To make it harder, the drive wasn’t empty: 2.5 TB of the 6 TB was already occupied when the test began.
I expected the cliff, but I got a staircase — with a really nice surprise at the end. Let me tell you all about it.
The setup
- Source: HP Z8 G4 workstation (Windows 11), internal storage (Samsung EVO 980 Pro, M2)
- Target: WD My Passport 6TB (SMR), connected via the supplied USB 3 cable to the fastest front-panel port (far right)
- Payload: 533 GB of render frames, ~200 MB each, in several folders — large, sequential, write-once data
- Drive state at start: ~2.5 TB already in use
- Start time: 10:00 AM
The results
| Phase | Data transferred | Sustained speed | Notes |
|---|---|---|---|
| 1 | 0–10 GB | ~100 MB/s | Rock solid, no fluctuation |
| 2 | 10–100 GB | 75–80 MB/s | Minor wobble, never below 75 |
| 3 | 100–~450/500 GB | 40–50 MB/s | Remarkably stable for hours |
| 4 | Final ~30–80 GB | 10–20 MB/s | Dips to 4 MB/s, occasional stalls |
Total: 533 GB in roughly 3.5 hours (10:00 AM – ~1:28 PM) — an overall average of about 42 MB/s, cache exhaustion included.
What surprised me
The drop everyone warns you about — the one I expected somewhere around the 100 GB mark at the latest — didn’t arrive until the transfer was 85–90% complete. The drive sustained 40–50 MB/s for nearly half a terabyte of continuous writing, on a drive that was already almost half full. Only in the final stretch did it become noticeably, annoyingly slow, with speeds dipping into single digits and the occasional stall.
Why a staircase and not a cliff
The explanation, I think, is that this workload is SMR’s best case — and that matters more than the technology’s reputation suggests.
The shingling penalty really bites on random writes and rewrites, where the drive must read-modify-write entire overlapping zones to change a few blocks. But large sequential files can be streamed more or less directly into shingled zones with comparatively little penalty. Mapping that onto the phases:
- Phase 1 (~100 MB/s): writes land in the drive’s CMR-managed media cache at full speed.
- Phases 2–3 (75–80, then 40–50 MB/s): the drive writes sequentially into shingled zones — something it can genuinely sustain — while juggling zone allocation on a drive that’s already 2.5 TB full.
- Phase 4 (10–20 MB/s, stalls): the final bill comes due. Hours of deferred media-cache cleaning debt catch up; the staging area is exhausted, and every incoming write now competes with real-time reshingling and eviction. The drive is doing two jobs at once with no buffer left — hence the stutter-and-stall pattern.
The remarkable part isn’t that the slowdown happened. It’s that the drive held it off for so long. Honestly, that little WD My Passport Drive exceeded my expectations (and I’m not being paid to say this).
But can you get the data back off quickly?
Fair question — a slow write penalty would be forgivable, but a slow read penalty would make the drive a data prison. So I measured that too: barely an hour after the transfer finished, I copied the same 533 GB back off the Passport onto an SSD. For the vast majority of the transfer, the drive read back at a steady 85–110 MB/s (measured per-disk with iostat — more on that in a moment): no staircase, no stutters. Shingling penalises writes, not reads; once your data is settled, it comes back at full sequential speed.
With one interesting footnote: the very last 40–50 GB of the read-back slowed down noticeably — and it was the same data that had been written during the drive’s cache-exhaustion phase, barely an hour earlier. My theory: files written while the drive was in crisis mode landed fragmented across media-cache remnants and hastily allocated zones, and the drive likely hadn’t finished its background tidy-up before I asked for them back. So the write penalty appears to leave a temporary scar on the most recently stressed data — one that should heal once the drive has had some idle time to finish reshingling. I’ll re-test those folders after a few days and report back.
A measurement tip while we’re here: I did the second transfer on my Mac Studio running Tahoe, and Activity Monitor shows aggregate disk throughput, and Spotlight indexing the destination SSD briefly convinced me the Passport was reading at 250+ MB/s — physically impossible for a 2.5″ drive. Per-disk iostat -d -w 5 diskN told the truth: 85–110 MB/s. Trust per-device numbers and wall-clock arithmetic, not pretty graphs.
A real-world data point to back up the read story: some footage I had previously offloaded to this drive needed checking in DaVinci Resolve. My project files still pointed at the internal drive, so I relinked the media against the Passport instead. After a moment of rejigging, Resolve played the 4K footage almost fluidly straight off the external drive — the occasional skip, but entirely usable for reviewing the material, and scrubbing back and forth worked fine too. For an archive drive that’s more than good enough: you can inspect what you’ve stored without copying it back first.
The honest caveat
This test says nothing about random writes, incremental backups full of small files, or rewrite-heavy workloads. In those scenarios, SMR would still suffer, and suffer badly — the reputation exists for a reason. If you’re planning to use an SMR drive as a working drive, a Time Machine target, or anything with constant small modifications, the cliff is still waiting for you.
A concrete counter-example from my own shelf: had I copied a batch of uncompressed Unreal Engine projects instead — thousands upon thousands of tiny files — this article would have told a very different, much uglier story. Same drive, same port, same cable; entirely different outcome. The workload is the variable, not the drive.
Give SMR some credit
The internet’s verdict on SMR — including, probably, some of my own past grumblings — is that these drives are categorically bad. That’s simply incorrect. They solve a problem we all have: more capacity at a lower price point. There is a penalty to pay for it, but it’s a penalty with a shape, and once you know the shape, you can work around it. If you’re storing video footage, render output, disk images, or any other large sequential files, you’ll have exactly the experience documented above — and you’ll have paid noticeably less per terabyte for it.
A practical tip: zip before you transfer
If you do find yourself needing to move a small-file monster onto an SMR drive — that uncompressed Unreal Engine project, a node_modules folder, a decade of email archives — don’t transfer it as-is. Zip it up into one large file first, then transfer the archive. Your drive will thank you for it.
Yes, the compression takes time — but it runs in the background on your fast internal storage while you get on with something else. What lands on the SMR drive is then a single large sequential file: exactly the workload this article shows these drives handle well. Don’t expect thousands of tiny files to land quickly, because they won’t; one big ZIP will get you a dramatically better transfer result. As a bonus, a single archive is also friendlier to copy back off the drive later.
The verdict
For its actual intended use case — a write-once archive for large files — the My Passport 6TB acquitted itself far better than its reputation suggests. 533 GB of render output at an average of ~42 MB/s, with the first 100 GB flying past at 75–100 MB/s, is a perfectly respectable result for a portable, bus-powered drive — and everything reads back at full speed afterwards, once the drive has finished its housekeeping.
SMR is fine, actually — if your workload is big sequential files. Know your workload, and the cliff becomes a staircase.
- Buy this drive through my affiliate link and support my projects: https://amzn.to/3TE27y6