Reading drive health: turning SMART and NVMe counters into a decision

Verified 2026-10-292 min read

Which attributes actually predict failure, how Windows exposes them, and the thresholds that mean replace instead of monitor.

Tested onPro (26H2) / Server · Build 26300.9457 / 26100.33650 · x64

Build: 26300.9457 / 26100.33650
Edition: Pro (26H2) / Server
Architecture: x64
Support: Documentation (E1)
Evidence: E1Official documentation
Verified: 2026-10-29
Page type: Method

The short answer

Health is not one number. Three signals matter, in this order:

Reallocated / pending sectors (HDD) or media errors (NVMe) — actual data loss risk.
Percentage used / endurance consumed (SSD) — how much write life is left.
Temperature and power-on hours — context, not a verdict.

A drive with zero reallocations and 90% endurance used is fine. A drive with three pending sectors and 10% used is a backup-then-replace.

What Windows itself reports

Get-PhysicalDisk | Select-Object FriendlyName,MediaType,HealthStatus,OperationalStatus,Size
Get-Disk | Select-Object Number,FriendlyName,HealthStatus,OperationalStatus
Get-CimInstance Win32_DiskDrive | Select-Object Model,InterfaceType,Size,Status,FirmwareRevision

HealthStatus is coarse: Healthy, Warning, Unhealthy. Windows flips to Warning only when the firmware reports a predictive failure — a drive can be dying for months at Healthy.

For anything beyond that, you need the SMART / NVMe log pages, which Windows does not surface in a supported UI. Vendor tools and third-party readers query them through IOCTL_STORAGE_QUERY_PROPERTY and the NVMe pass-through interface.

The attributes worth reading

HDD (ATA SMART):

Attribute	Meaning	Act when
05 Reallocated sectors	sectors remapped to spares	any non-zero, rising
C5 Current pending sectors	unreadable, awaiting remap	any non-zero
C6 Uncorrectable	could not be read even after retry	any non-zero
09 Power-on hours	age	context only
C2 Temperature	thermal history	sustained above spec

SSD (NVMe):

Field	Meaning	Act when
Percentage used	endurance consumed	>80% start planning replacement
Media and data integrity errors	uncorrectable errors	any non-zero
Available spare	spare blocks remaining	below threshold (often 10%)
Critical warning	firmware-level flags	any flag set
Temperature	controller temp	near throttle point

A reproducible workflow

Snapshot the numbers so you can compare later — one reading proves nothing.
Re-read after a week. A stable pending sector is different from a rising one.
Check the event log for storage errors that Windows already noticed:

Get-WinEvent -FilterHashtable @{LogName='System'; ProviderName='disk','Ntfs','volmgr','stornvme'} -MaxEvents 30 |
  Select-Object TimeCreated,ProviderName,Id,LevelDisplayName,Message

Check the filesystem if you suspect corruption:

Repair-Volume -DriveLetter C -Scan

-Scan is read-only. Do not run a repair on a drive you have not backed up.

Decision table

Evidence	Decision
Pending/reallocated sectors, any count	Back up now, replace soon
Media errors on NVMe	Back up, plan replacement
Endurance >80%	Start planning; keep backups current
`HealthStatus : Warning`	Replace, do not “monitor”
Only high hours, clean counters	Keep using; keep backups
SMART unreadable behind RAID	Use the controller’s own health view

Why “SMART passed” is not a warranty

SMART predicts some failures and misses others. Controllers die, firmware bugs brick drives, and some drives fail with no warning at all.

The only reliable protection is a backup you have restored at least once. Health monitoring tells you when to hurry.

Lab boundary

E1. Attribute semantics come from the ATA/NVMe specifications and vendor documentation. The lab’s virtual disks do not expose real SMART data, so no failing drive was observed.