People treat SSDs like milk. Three years, toss it. That’s HDD folklore with a new label. A consumer drive has a write budget (TBW, terabytes written) printed in the spec. A 1TB TLC disk used as a normal Windows box might see 10–20 TB a year. The box on the Amazon listing often says 300–600 TBW. Math says a decade-plus, often more. What I actually replace is not “it hit 100% wear.” It’s a controller that quit, a mapping table that ate itself after a power yank, or SMART that started throwing caution while Windows still “felt fine.”
What “lifespan” even is
NAND cells take a finite number of program/erase cycles. The controller spreads writes around (wear leveling) and keeps spare blocks in its pocket (overprovision). TBW is the maker’s bet on that. Warranty years are a lawyer number, not a fuse. I’ve seen five-year drives die in month eight and three-year QLC still green at 4% wear. The spec sheet is a ceiling, not a countdown.
| Kind of disk (rough) | What I use it for | What wears it out faster |
|---|---|---|
| TLC (most mainstream 1TB NVMe) | OS, games, normal work. This is the default I buy. | Constant downloads, CCTV, scratch disks, moving huge files all day. |
| QLC (cheap big drives) | Cold storage, a second game library. Fine if you’re not rewriting it. | Same as above, only sooner. Writes get ugly once the fast cache is full. |
| DRAM-less (HMB) budget NVMe | Browser laptop. I don’t put a video project on one. | Heavy random writes. Not automatically “dead in a year.” Just not a work disk. |
I don’t memorize every TBW. CrystalDiskInfo (or the maker’s tool) is where I look. Percentage Used counting up from 0 is NVMe wear. Total Host Writes is how much you actually threw at it. Available Spare is the other line. That one dropping is a sick pool, not a birthday. SATA sometimes says Media Wearout and the number runs the other way, 100 down toward 0, so don’t treat a 3 there like 3% used. Percentage Used at 3% after two years of home use is boring. That’s good. 90% on a torrent box is the budget doing what you asked; still copy off before it hits Caution. 40% in six months on a laptop that only does Chrome is a bad stick or a firmware that’s counting wrong. Clone anyway. Don’t argue with SMART.
Signs I trust (SSD, not a clicking HDD)
SSDs don’t click. If you hear a click it’s the old spinning disk, a fan, or a coil. Don’t sit there waiting for a death rattle that never comes.
| What you see | What I think | What I don’t do |
|---|---|---|
| SMART Caution/Bad, spare blocks dropping, uncorrectable / integrity errors climbing | Clone today. Wear or a sick controller. Either way the data is the job. | Defrag, CHKDSK “to heal it,” or a 12-hour surface scan. |
| Drive goes read-only. Copy off works, install fails, Windows says the disk is write-protected | A lot of dying SSDs do this on purpose. Last-gasp mode. Clone off. That disk is done as a boot drive. | A format “to clear the lock.” |
| Vanishes in BIOS, then comes back, then gone again | Controller or power to the M.2. Power off, reseat once. If it repeats, clone while it still enumerates. | A new motherboard first. |
| Writes crawl after a few gigs, then recover after a rest | SLC cache dumped into slow NAND. Common on cheap QLC when the disk is full. Annoying. Not the same as SMART Bad. | Don’t replace it on that alone if SMART is still good. Free up 20% and see if writes recover. |
| Sudden “0 bytes free” or the capacity in Disk Management looks tiny | Firmware or a partition mess. Sometimes a dying map. Don’t keep writing to it. | A “cleaner” app. |
| Event Viewer disk errors, random BSOD with storage in the stack, files that corrupt after a copy | Treat as dying until SMART and a clone say otherwise. | Reinstall Windows on the same stick as the fix. |
Signs I ignore
- The warranty sticker year. I read SMART, not the box.
- It’s an SSD so it “must” be faster than this HDD laptop. A full QLC NVMe, hot under the shield sticker, will feel drunk. Thermals and a full disk. Not dying.
- One CrystalDiskMark run that’s slower than a YouTube review. Bench the same disk twice. Laptops throttle. USB enclosures lie about speed.
- Windows “optimize” / TRIM looking busy. That’s maintenance. Not a death rattle.
- 80°C in a game for ten minutes, then it drops. A lot of NVMe will do that under load and throttle. I don’t treat it as a funeral. If it sits at 80 idle, then I look at the sticker / heatsink / a dying fan. See the hot-laptop guide.
How I actually check
CrystalDiskInfo on Windows is enough for a first look. Health, temperature, Percentage Used, Available Spare, total writes, and the caution flags. The vendor toolbox (Samsung Magician, WD Dashboard, whatever’s on the stick) is better for firmware. I only run that when the drive is still healthy. A firmware update on a Caution disk is how you brick the last copy.
- Screenshot SMART before you do anything. You’ll want that if it dies mid-clone.
- If BitLocker or device encryption is on, suspend it before you clone. Same note as the SSD upgrade guide.
- A USB enclosure is for cloning and for seeing whether a “dead” NVMe enumerates on another machine. One data point. Enclosures have their own failures.
On the bench
USB NVMe / SATA enclosure (clone and test, not a speed toy)
Same box I use in the upgrade guide. If SMART is ugly, this is how the data leaves. If the laptop won’t see the stick, this is how I find out whether the SSD or the slot is the liar.
- USB 10Gbps is enough for a clone.
- Don’t trust enclosure temps or benches as gospel.
When I buy a new one
The replacement OS disk. After the copy exists. Not before. A mainstream 1TB TLC NVMe for an OS disk. I don’t chase PCIe 5.0 numbers for a laptop that will thermal-throttle a 3.0 drive. I don’t buy the cheapest 2TB QLC as a boot disk for someone who dumps 4K video on it all week. Same instinct as the upgrade guide: browser box can be cheap, work disk isn’t. This one stops at: prove it’s dying or worn, copy off, then shop.
On the bench
A known 1TB TLC NVMe (after the copy is safe)
Replacement, not a trophy. Match the slot length (2280 vs 2242). If the old one died at 8% wear, I don’t cheap out on mystery QLC for the OS disk.
- Clone onto it from the good copy, not from the dying stick if the dying stick is already flaking.
- Soldered eMMC/SSD in an ultrabook is not this job. If you can’t see a module, stop.
What I tell people who just want a number
Home OS disk, TLC, not full, not a download server: I expect years, and I expect the failure mode to be sudden (gone / read-only) more often than a polite 100% wear bar. Heavy write box: read SMART twice a year. Don’t wait for it to feel slow. Slow is late. And if it’s already caution, that’s not a lifespan essay. That’s clone-first.
Verdict
Read SMART. Don’t retire a healthy SSD because the calendar said so. Read-only, vanishing in BIOS, spare blocks dropping: copy off the same day. Then buy the disk. Shopping while the only copy is the sick one is how this ends.