NVMe SMART health explained: percentage used, media errors and unsafe shutdowns

An SSD utility that says “Health: 97%” looks reassuring, but it hides the fields that explain what is actually happening. NVMe defines a SMART / Health Information Log with warnings, spare capacity, endurance use, temperature, written data, power history and error counters. The useful diagnosis comes from reading those fields together.

SMART is also not a backup. A healthy log cannot protect against accidental deletion, controller failure or a sudden event the firmware did not predict.

Read Critical Warning first

The NVMe Critical Warning field is a bitmask, so more than one condition can be active. The standard includes warnings for [1]:

• available spare below its threshold;
• temperature above or below a critical threshold;
• degraded NVM subsystem reliability;
• media placed in read-only mode;
• failure of the volatile-memory backup device, where present.

A non-zero Critical Warning is more actionable than a rounded health score. Back up important data first, then identify which bit is set. If the drive has entered read-only mode, the controller may be protecting the remaining readable data rather than behaving like a normal writable device.

Percentage Used is life consumed, not life remaining

NVMe defines Percentage Used as a vendor-specific estimate of the percentage of NVM subsystem life consumed, based on actual use and the manufacturer's prediction [1]. Two details matter:

1. 100% may not indicate failure. It marks estimated endurance consumption, not an instant physical death.
2. The value may exceed 100. The specification allows reporting beyond the rated point.

This is why a utility's “health” display can confuse buyers. Some tools convert Percentage Used into a remaining-life number such as 100 - used. That is a convenient presentation, not a universal promise of exact days left.

Use Percentage Used for planning:

• trend it over time;
• compare it with Data Units Written and the drive's rated TBW/DWPD;
• forecast when the rated endurance point will be reached;
• replace before the risk becomes operationally unacceptable.

For endurance math, see TBW and DWPD explained.

Available Spare is a separate signal

Available Spare is a normalized percentage of remaining spare capacity. The log also reports an Available Spare Threshold. If spare falls below that threshold, the corresponding Critical Warning bit is set [1].

Do not confuse this with free space in the filesystem. Available spare belongs to the SSD controller's internal pool for replacing unusable NAND areas. Deleting user files does not refill worn-out physical spare blocks.

A falling spare value, especially with media errors, is a stronger reliability concern than ordinary capacity usage.

Media errors and error-log entries are not the same counter

The SMART log exposes Media and Data Integrity Errors and Number of Error Information Log Entries. The first covers unrecovered data-integrity events; the second counts entries added to the NVMe error log and can include command-related errors [2].

Interpret them by trend and context:

An error counter is cumulative and may not reset after the cause disappears. The important question is whether it continues to grow under normal use.

Unsafe Shutdowns records events, not proven corruption

An unsafe shutdown is a loss of power or shutdown without the expected clean notification. The SMART log counts these events [2][4]. Common causes include a forced power-off, battery removal, unstable enclosure, surprise unplug or system crash.

The number does not say how many files were corrupted. A drive may recover every time, or one interruption may hit a critical metadata update. Treat the field as evidence about the power environment:

• compare the increase with known outages;
• check whether media errors or filesystem repairs followed;
• inspect USB/NVMe enclosure and sleep behavior;
• use storage with suitable power-loss protection when the workload cannot tolerate interrupted writes.

Our power-loss protection guide explains the difference between protecting data at rest and data in flight.

Temperature needs a trend, not one screenshot

The health log includes composite temperature, warning/critical temperature time and, where exposed, sensor readings and thermal-management transitions [3][4]. A single high value after a benchmark is less informative than repeated time above the warning threshold or frequent thermal throttling.

Check:

• idle and sustained-load temperature;
• time above warning and critical thresholds;
• airflow, heatsink contact and enclosure temperature;
• whether performance drops at the same time temperature rises.

Use the drive maker's specified operating range and thresholds. Do not apply one internet “safe temperature” to every controller and form factor.

A practical reading order

When reviewing an NVMe SMART report:

1. Back up if the data matters and any warning is active.
2. Decode Critical Warning bits.
3. Check Media and Data Integrity Errors and detailed error-log trend.
4. Compare Available Spare with its threshold.
5. Read Percentage Used and Data Units Written against rated endurance.
6. Review warning/critical temperature time, not only current temperature.
7. Correlate Unsafe Shutdowns with host power history.
8. Record the report with date, firmware and workload so the next reading has context.

Bottom line

NVMe SMART is a set of evidence, not one health percentage. Critical warnings, media errors, spare capacity and read-only status tell you about immediate risk. Percentage Used tells you how much rated endurance the manufacturer estimates has been consumed. Temperature, written data and unsafe shutdowns explain the operating history. Read them together, trend them, and replace a drive according to the consequence of failure — not a colorful 97% badge.