How to Read Drive SMART Values and Warnings

The short answer: A single SMART row contains three different numbers—the raw value (what the drive actually measured), a normalized value from 1 to 254, and the manufacturer’s threshold. An attribute has only “failed” when its normalized value falls to or below that threshold; the Pre-fail label means nothing by itself. For hard disk drives, five attributes matter most: 5, 187, 188, 197, and 198. For NVMe drives, focus on Percentage Used, Available Spare, and Media and Data Integrity Errors—and only these have consistent definitions in the specification. What matters most is not a single reading but the difference between two measurements.

This article is a glossary, not a procedure. For what to do about bad sectors, when to shut down the computer immediately, and how cloning works, see HDD bad sectors and when to shut down a drive with bad sectors. Here, we focus solely on reading the table.

One Row, Three Different Numbers

According to the smartctl manual, each attribute has a raw value (RAW_VALUE) and a normalized value (VALUE). Each manufacturer uses its own algorithm to calculate the normalized value from the raw data, producing a number from 1 to 254. There is also a threshold (THRESH) and a WORST column showing the worst recorded state.

There is only one rule: if the normalized value is less than or equal to the threshold, the attribute has failed. Smartctl displays this as FAILING_NOW in the WHEN_FAILED column. If the failure occurred previously and the value later recovered, it displays In_the_past.

Two columns cause the most confusion:

• TYPE — either Pre-fail or Old_age. The manual is unambiguous: an attribute being classified as Pre-fail does not mean the drive will fail. It only has that meaning if the normalized value has fallen to the threshold. Failure of an Old_age attribute indicates that the designed service life has been exhausted, not that a failure is imminent.
• UPDATED — Always for values updated during normal operation and Offline for those refreshed only during an offline test. Values in the second group may be outdated.

One point explains most arguments on discussion forums: since revision 4 of the ATA/ATAPI-4 standard, the meaning of attribute fields has been entirely up to the manufacturer. There is therefore no universal “table of correct values” that applies across all brands.

Five Values That Matter on a Hard Drive

A Backblaze analysis from October 2016 draws on data from its own drive fleet and identifies five attributes for which a raw value above zero warrants checking the drive. In CrystalDiskInfo, you need to look for their hexadecimal labels because both attribute IDs and raw values are displayed in hexadecimal format—which is why you see C5 instead of 197.

Backblaze provides two figures that only make sense side by side: at least one of the five attributes had a value above zero in 4.2% of operational drives but 76.7% of drives that failed. The difference between those percentages represents the practical value of SMART. The remaining figure is the warning: 23.3% of failed drives had zeros in all five attributes. That is why the only genuine safeguard is a backup following the 3-2-1 rule, not a color-coded diagnostic status.

According to Backblaze, attributes 197 and 198 are strongly correlated. It monitors both because the correlation is not perfect and not every manufacturer reports both.

False Alarms: What Does Not Mean Trouble Yet

According to the CrystalDiskInfo documentation, the program reports Caution once the raw value crosses its threshold. On hard drives, attributes 05, C5, and C6 trigger this status. On SSDs, it appears when the remaining lifetime reaches 10% or less. It therefore means “this value is no longer zero,” not a prediction of when the drive will fail.

Common false alarms include:

• a single Command Timeout (188) after an unexpected shutdown, a faulty cable, or an overloaded adapter—a count that does not increase for a year;
• a few reallocated sectors (5) that remain unchanged for years—the drive retired several locations and continues operating with its spare capacity;
• a normalized value of 100 for an attribute with a threshold of 0—such an attribute cannot “fail”; the manufacturer uses it only as a counter;
• values from a device connected through a USB bridge that does not pass SMART commands through—the program may then show nonsense or nothing at all.

Four situations, however, are not false alarms: a Current Pending count that increases between two readings, any non-zero value for attribute 198, attributes 5 and 197 occurring together, and an attribute marked FAILING_NOW. The first three mean that the drive has already failed to read data it previously wrote itself.

SATA SSDs: The Same Table, a Different Language for Every Manufacturer

Reading SMART data from a SATA SSD is the most difficult case, and it is not your fault. In a June 2023 analysis of SSD attributes, Backblaze compared three models from different brands—Seagate, Western Digital, and Crucial—and found that only five of 44 attributes were shared. That is 11%, compared with 42% among the hard drive models examined. Participation in the standard is voluntary, so manufacturers define their own names and meanings.

In practice, this means you should never compare absolute values across SATA SSD brands. Focus on values documented by the manufacturer—typically remaining life, wear leveling (attribute 173), and unexpected power loss (174). Place more trust in the manufacturer’s utility and changes over time than in a generic attribute table. Symptoms that SMART may not detect at all are covered in signs that an SSD is dying.

NVMe: Values with Consistent Definitions

This is one area that has genuinely improved in recent years. NVMe does not use freely interpreted attributes. Instead, it provides log page 02h with consistently defined fields—the same names and units across manufacturers. The current format is described in revision 2.3 of the NVM Express Base Specification, dated August 1, 2025; the complete NVMe 2.4 specification set was released on August 4, 2026.

Two details in the specification change how “wear” should be viewed. Percentage Used is a manufacturer’s estimate and is updated once per power-on hour, so it will not move immediately after a short write operation. A value of 100 means that the estimated endurance has been exhausted but, according to the specification, does not necessarily indicate a failure. For endurance measurements, the specification refers to JEDEC standard JESD218B-02.

In practical terms, Percentage Used is a purchasing-planning indicator. Available Spare approaching zero and Media and Data Integrity Errors above zero are data-safety indicators.

How Much Data You Have Actually Written

For an SSD, the total amount written provides the fairest comparison with its rated endurance, but it is often misread. Under the specification, Data Units Written uses units of 512 bytes, but the value is reported in thousands and rounded up. A value of 1 therefore represents 1,000 units of 512 bytes each.

The conversion is simple: multiply the value by 512,000 to obtain bytes. A value of 2,500,000 therefore means 1.28 TB written. Fortunately, smartctl performs this conversion automatically and displays the resulting volume in parentheses beside the raw number. This is the figure you should compare with the manufacturer’s TBW endurance rating for your specific model.

Tools for Reading SMART Data

• CrystalDiskInfo — displays the full attribute table, with IDs and raw values in hexadecimal, plus the relevant log-page values for NVMe drives.
• smartctl from the smartmontools package — displays the same values as text; for NVMe drives, it reads attributes from the SMART/Health Information log. It is the best choice when you want to save the output and compare it a month later.
• PowerShell without additional software — according to Microsoft’s documentation, Get-PhysicalDisk | Get-StorageReliabilityCounter | Format-List can display values such as Wear, PowerOnHours, Temperature, and counts of uncorrected errors. Only fields reported by both the drive and its driver are populated; the Wear field is empty in Microsoft’s documentation example.
• the manufacturer’s utility (Samsung Magician, Crucial Storage Executive, or WD Dashboard) — often the only source that correctly explains proprietary attributes on SATA SSDs.

Reading the table once does not place a meaningful load on the drive. Save the output with the date; a second measurement a month later will tell you more than any online threshold table.

When Reading the Values Is No Longer Enough

The distinction between the two possible outcomes is straightforward. If wear or the reallocated-sector count is increasing but all data remains readable, the solution is drive replacement and system migration. This is routine service work, starting at approximately 600 Kč plus the cost of the part, depending on the scope. Current rates are listed in our price list and on the SSD replacement page. Once attribute 198 or Media and Data Integrity Errors rises above zero, the issue becomes data recovery, and the opposite rule applies: the less the drive is used, the better.

At our Brno service center, we can read SMART data from both hard drives and NVMe drives, compare the values with the computer’s actual behavior, and tell you which of those two categories your drive falls into. Reading and assessing the values is treated as diagnostics—starting at approximately 600 Kč if no repair is performed, according to the price list. If a repair is carried out, diagnostics are handled as part of the job; the exact price always depends on the drive type and scope. Bring information about your latest backup as well: SMART gave no warning at all for 23.3% of the failed drives in the Backblaze analysis, so a clean table is an argument for having a backup, not against it.