S.M.A.R.T.: drive health report

Your SSD keeps a diary. It logs temperature, errors and wear, and it never complains until you ask to read it.

The short answer

S.M.A.R.T. (Self-Monitoring, Analysis and Reporting Technology) is the monitoring system built into most ATA/SATA and SCSI/SAS hard drives and SSDs. Its purpose is to watch the drive's reliability, predict failures and run self-tests [1]. NVMe drives offer the same idea as the "SMART / Health Information" log page [3][4]. Our assessment: a "PASSED" verdict is a hint, not a guarantee, so read the counters and keep a backup either way.

How it works

1. SATA drives report attributes. Each attribute has an ID, a raw value, a normalized value (VALUE), the lowest value seen (WORST) and a threshold (THRESH). The drive's firmware converts raw to normalized, and smartctl only reports what it reads [1]. If the normalized value is less than or equal to the threshold, the attribute has failed. "Pre-fail" is just the attribute type and does not mean the disk is about to fail [1]. The attribute table was never part of the ATA standards, and even the general format was removed from the standard with ATA-4 in 1998, so meanings differ by vendor [2].
2. The health verdict. smartctl -H prints it. A failing status means the drive has either already failed or predicts failure within 24 hours, and the man page says to get your data off the disk as soon as you can [1]. For NVMe, the verdict comes from the "Critical Warning" byte of the SMART/Health log [1].
3. NVMe has its own log. Log page 02h holds Critical Warning bits (spare below threshold, temperature, reliability degraded, read-only, backup device failed), Available Spare with its threshold, and Percentage Used [3][5][6]. Percentage Used is a vendor-specific estimate. A value of 100 means the estimated endurance is consumed, which may not mean failure, and the value may exceed 100 [3]. Data Units Read and Written count in thousands of 512-byte units [5]. The log also has power-on hours, unsafe shutdowns and media errors [3][6]. SATA SMART uses a different attribute list [6].
4. Self-tests. A short test usually takes under ten minutes, an extended test tens of minutes to several hours. Results land in the self-test log, and a running test can slow the drive [1].
5. Know the blind spots. A USB bridge or RAID controller can hide the real result. smartctl then warns and falls back to checking pre-failure attributes against their thresholds [1].

A practical example

Run these as root or with sudo. For NVMe, -a equals -H -i -c -A -l error -l selftest, and the device is /dev/nvme0 [1]:

sudo smartctl -H /dev/sda            # verdict (SATA)
sudo smartctl -A /dev/sda            # attribute table
sudo smartctl -a /dev/nvme0          # NVMe: verdict, health log, error log
sudo smartctl -t short /dev/sda      # start a short self-test
sudo smartctl -l selftest /dev/sda   # read the result later

Say an NVMe drive shows Critical Warning 0x00 and Percentage Used 3. That is a healthy-looking reading, but those numbers are made up for illustration. Our inventory (as of 14 to 16 Oct 2026) does not record SMART values of our own drives, so we claim none.

Checklist

• Back up first, then test. If the verdict is failing, copy your data off before anything else [1].
• Read the full table, not only the verdict.
• Compare each normalized value with its threshold, and skip raw values you cannot interpret [1][2].
• Check Critical Warning, Available Spare against its threshold, Percentage Used and media errors on NVMe [3][5].
• Write down the values with the date, so you see a trend instead of a single snapshot.
• Try -d sat when an ATA disk sits behind a USB bridge, and read the controller's manual for RAID [1].
• Keep a 3-2-1 backup regardless; see our entry on the 3-2-1 backup rule.