Five SMART numbers that mean your drive is dying

Every modern drive reports SMART data — dozens of counters, most of which tell you nothing useful. Backblaze, which monitors over 341,000 drives, acts on just five. If any of them rises above zero, the drive gets replaced.

The five that matter

ID	Attribute	What it means
5	Reallocated Sectors Count	Sectors that failed and were remapped to spares. The classic sign of surface degradation.
187	Reported Uncorrectable Errors	Errors the drive could not fix with its own error correction. Data was at risk.
188	Command Timeout	Operations that timed out, often signalling electrical or interface trouble.
197	Current Pending Sector Count	Sectors failing to read and queued for reallocation.
198	Uncorrectable Sector Count (offline)	Sectors that could not be read even during offline scanning.

Backblaze's own analysis found that 76.7% of failed drives had at least one of these five above zero, while only 4.2% of healthy, operational drives did. That combination — catching three quarters of failures with few false alarms — is why these five are worth watching and the rest largely are not.

The uncomfortable part: a clean report proves nothing

The same analysis found 23.3% of failed drives gave no SMART warning at all. Google's landmark study of its own fleet, Failure Trends in a Large Disk Drive Population (2007), put that figure even higher, at 36% of failed drives. Google's conclusion was blunt: models based on SMART parameters alone are unlikely to be useful for predicting individual drive failures.

So SMART is a good early-warning system and a poor guarantee. Between a quarter and a third of drives simply die without notice.

Two myths the data kills

“Keep drives as cool as possible.” Google found the correlation between temperature and failure surprisingly weak, and across the range they studied lower temperatures were associated with higher failure rates at most drive ages. Within normal operating ranges, temperature is not the lever people think it is.

“Spin drives down to save them.” Power cycling is a real stressor. When Backblaze investigated a mechanical fault in Q1 2026, one of its mitigations was explicitly to reduce power-cycle frequency — a good argument for leaving a NAS spun up rather than aggressively idling it.

How to check your own drive

Windows: CrystalDiskInfo — free, and shows raw values for all five attributes.
macOS / Linux: smartctl -a /dev/sdX from the smartmontools package.
NAS: Synology, QNAP and TrueNAS all expose SMART data and can email you when it changes.

Any non-zero value in SMART 5, 187, 188, 197 or 198 means: back up now, then plan a replacement. On a drive you have just received — particularly a used, renewed or warehouse one — it is grounds for an immediate return.

Sources: Backblaze, What SMART Stats Indicate Hard Drive Failures; Pinheiro, Weber and Barroso (Google), Failure Trends in a Large Disk Drive Population, USENIX FAST 2007.

Find a replacement drive Failure rates by model

Frequently asked questions

Which SMART attributes actually predict drive failure?

Backblaze acts on five: SMART 5 (Reallocated Sectors Count), 187 (Reported Uncorrectable Errors), 188 (Command Timeout), 197 (Current Pending Sector Count) and 198 (Uncorrectable Sector Count, offline). Across its fleet, 76.7% of failed drives had at least one of these above zero, against only 4.2% of healthy drives.

My SMART data is clean. Does that mean my drive is healthy?

Not reliably. Backblaze found that 23.3% of failed drives gave no SMART warning at all, and Google's large-scale study put the figure at 36%. SMART is a useful early-warning system, not a guarantee, so a clean report is never a substitute for backups.

Does running a hard drive cooler make it last longer?

Within normal operating ranges, the evidence says no. Google's study of its own fleet found the correlation between temperature and failure much weaker than assumed, and over the range studied lower temperatures were associated with higher failure rates at most drive ages. Keeping a drive inside its rated range matters; chilling it below that does not appear to help.