The short answer: A single number in CrystalDiskInfo is not a diagnosis. The Samsung 990 Pro with the original 0B2QJXD7 firmware reported its „health„ dropping by tens of percentage points after just a few terabytes of writes, but this was a firmware bug. The Crucial MX500 triggers „pending sector" alerts even though Micron uses this attribute to count ECC events. A WD Green with IntelliPark can rack up hundreds of thousands of head-parking cycles simply because its factory timer is set to eight seconds. The reverse is also true: according to Backblaze data, almost a quarter of drives failed without any of the five key attributes ever moving above zero. Base your decisions on several appropriately weighted attributes, their trends, and the drive’s actual behavior—not on a percentage displayed on the screen.
We explain how to read SMART, which four attributes to monitor, and what a dying drive sounds like in How to Tell Whether an HDD or SSD Is Dying. This article follows on from that one and looks at situations where those same attributes can be misleading. We cover actual Seagate power-on hours from the FARM log and NVMe temperatures separately.
False Alarm: SMART Reports a Fault That Is Not There
Samsung 990 Pro: Wear Invented by Firmware
The drive launched in November 2022 with firmware 0B2QJXD7. Within weeks, reports began mounting that Percentage Used was rising and „health„ in Samsung Magician and CrystalDiskInfo was falling at an absurd rate. According to the Rossmann Group’s summary, users lost 10–30% within days after writing less than 2 TB—on a 2TB model rated for 1,200 TBW. Samsung acknowledged that the „anomalies were caused by firmware issues" and released firmware 1B2QJXD7 through Magician on February 13, 2023.
What the fix does not do is important: SMART values do not return to their factory state after the update. A drive that had already fallen to 78% stays at 78% and ages normally from there. The percentage on such a drive is therefore not a measurement of NAND wear but a record of a software bug. Other values matter more on the 990 Pro: the firmware version (smartctl -i or Magician), non-zero Media and Data Integrity Errors, Critical Warning, and Available Spare below its threshold. If those entries are clear, a low percentage left behind by the old bug is not a reason to replace the drive. And because even a supported update may, according to Tom’s Hardware, fail and render the drive unusable in a small percentage of cases, back up your data before flashing the firmware.
Crucial MX500: A „Pending Sector" That Is Not a Sector
A typical scenario looks like this: Proxmox, TrueNAS, or smartd sends an email saying „1 Currently unreadable (pending) sectors." An hour later, another message says the value has returned to 0, and the cycle repeats a few days later. On the Proxmox forum, the owner of a CT1000MX500SSD1 describes exactly this behavior, even though a manual smartctl check showed no bad sectors.
The explanation is in the manufacturer’s documentation. In technical note TN-FD-22, Micron defines attribute 197 (C5h) as Current Pending ECC Count: the raw value is the „total number of ECC events detected during host commands (such as READ) or background operations.„ Both the current and worst normalized values are fixed at 100, so this attribute can never push the drive below its threshold. It is not a sector waiting to be remapped, but a counter of corrected errors that may appear and then disappear. Tools that automatically translate attribute 197 as „pending sector" turn it into a critical fault.
The smartmontools project handles this in its drive database: the entry for Crucial/Micron client SSDs (MX100 through MX500 and BX500) renames the attribute to Current_Pending_ECC_Cnt, noting that on the MX500 the value „may jump 0 ↔ 1" (ticket #1227). Updating the database with update-smart-drivedb stops the smartd alerts. Forum threads also mention firmware versions M3CR022 and M3CR023 as possible fixes, but we could not find an official changelog confirming this. On an MX500, monitor attributes 5 (Reallocated NAND Blocks) and 202 (Percent Lifetime Remaining). If 197 briefly changed while attribute 5 did not move, nothing happened.
WD Green with IntelliPark: Hundreds of Thousands of Parks Set by the Manufacturer
WD Green drives—and, according to user reports, some other WD product lines—have an „idle3" firmware timer that moves the heads off the platters and onto the parking ramp after eight seconds of inactivity. WD calls this feature IntelliPark. In Windows, a drive at rest really can remain idle. In Linux, a NAS, or any environment that continuously writes metadata, the system wakes it up again every few seconds. The Manjaro wiki describes 100 or more parking events per hour. StorageReview measured an increase of 600–800 cycles per day for a WD20EARS in a Synology NAS, while a user on the WD forum reported 1,200 cycles during the first 24 hours on a new WD Red WD30EFRX.
According to the WD Green specification covering models WD5000AZRX through WD40EZRX, the drive is designed for 300,000 load/unload cycles. At 100 cycles per hour, it reaches that figure after roughly four months of operation—and attribute 193, Load Cycle Count, then shows a number that monitoring tools may present as an approaching end of life. The Manjaro wiki describes a drive with 1,991,353 cycles after 29,549 hours that continued to work. Attribute 193 measures wear on the head-parking mechanism, not read errors; attributes 5, 197, and 198 tell you about the platters.
You can change the timer with the official DOS utility wdidle3 (/S300 sets it to 300 seconds, /D disables it) or with the idle3-tools package on Linux:
sudo idle3ctl -g /dev/sda # show the current idle3 timer value
sudo idle3ctl -s 129 /dev/sda # set it to 30 seconds
sudo idle3ctl -d /dev/sda # disable head parkingThe change takes effect only after the drive has been fully powered off and back on; a restart is not enough. The authors of idle3-tools describe the utility as experimental, and accessing low-level drive data is at your own risk—so back up your data first here as well.
False Confidence: SMART Is Silent While the Drive Is Dying
In 2016, Backblaze evaluated 67,814 drives in its data center. Of the drives that failed, 76.7% had a non-zero value in at least one of attributes 5, 187, 188, 197, and 198. The remaining 23.3% failed with zeros in all five. At the same time, 4.2% of drives still in service also had a non-zero value—these attributes indicate risk, not certainty, in either direction.
The reason is simple: SMART counts only what the drive itself notices while reading and writing. There is no counter for a weakening head, bearing, motor, or electronic component. A drive with clean SMART data that has started clicking, slowed to a fraction of its normal speed, or begun disconnecting under load has a mechanical or electrical fault, regardless of whether the tool says „Good."
The situation is even more pronounced with SSDs because their failure is usually caused not by the wear tracked by Percentage Used, but by the controller or firmware. In November 2019, HPE warned that its SAS SSDs with firmware older than HPD8 would fail after exactly 32,768 hours of operation. After failure, neither the drive nor its data could be recovered, and drives deployed at the same time would fail almost simultaneously—one administrator described six SSDs dying within 15 minutes. In 2023, portable SanDisk Extreme and Extreme Pro V2 drives (SDSSDE61-4T00 and SDSSDE81-4T00) began displaying the message „The disk you attached was not readable by this computer" without prior warning. WD promised a firmware fix in May, and a class-action lawsuit followed in August. We explain separately why data recovery from an SSD is a different discipline from HDD recovery in situations like these.
USB is a third source of false confidence. A drive in an external enclosure responds only if the bridge passes through ATA commands (SAT). For this purpose, smartctl supports -d sat as well as special device types for JMicron, Cypress, and Sunplus. Some JMicron bridges do not support 48-bit ATA commands, so the extended error log (-l xerror) is disabled. An NVMe drive in a USB enclosure can only be read through supported ASMedia, JMicron, and Realtek bridges (-d sntasmedia, -d sntjmicron, and -d sntrealtek). If a tool shows nothing or reports an „unknown status" for an external drive, that does not mean the drive is healthy—it means its SMART data is unreadable.
Which Attributes Matter
| Attribute | What It Counts | How to Interpret It |
|---|---|---|
| 5 Reallocated Sectors Count | sectors remapped to spare capacity | any increase = platter or head problem; one of Backblaze’s five attributes |
| 187 Reported Uncorrectable Errors | errors that ECC could not correct | Backblaze schedules replacement as soon as it rises above 0 |
| 188 Command Timeout | commands aborted because the drive timed out | should be zero; interpret increases alongside 5 and 197 |
| 197 Current Pending Sector Count | sectors awaiting remapping after an uncorrectable read | HDD: back up immediately; Micron/Crucial SSD: ECC counter |
| 198 Offline Uncorrectable | uncorrectable errors found during offline reads | carries the same weight as 197 |
| 199 UDMA CRC Error Count | transmission errors over the cable | cable, port, or controller—not the platters; the value itself does not decrease |
| 193 Load Cycle Count | head-parking events | wear on the mechanism, not a data error (IntelliPark) |
| NVMe Critical Warning | flags: spare below threshold, temperature, reliability, read-only mode | smartctl -H reads only this on NVMe |
| NVMe Available Spare / Threshold | remaining spare capacity and its threshold | below the threshold = drive is exhausting its reserves |
| NVMe Percentage Used | manufacturer’s estimate of consumed endurance | 100 does not necessarily mean failure and may exceed 100; incorrect on the 990 Pro with 0B2QJXD7 |
| NVMe Media and Data Integrity Errors | uncorrected integrity errors (ECC, CRC, LBA tag) | non-zero and rising = back up immediately |
Case Overview: What SMART Reports and What Is Actually Happening
| Model / Situation | What SMART Reports | Reality | What to Do |
|---|---|---|---|
| Samsung 990 Pro, fw 0B2QJXD7 | Percentage Used rises; health drops by percentage points within days | firmware bug, not wear | check firmware; after backing up, install 1B2QJXD7; base decisions on Media Errors and Critical Warning |
| Crucial MX500 | 197 = 1, then 0; smartd sends alerts | on Micron drives, 197 is an ECC counter | update the smartmontools drive database; monitor 5 and 202 |
| WD Green (idle3 set to 8 s) | 193 in the hundreds of thousands | head-parking timer; rated for 300,000 cycles | use wdidle3 / idle3ctl, then power off and back on; monitor 5/197/198 |
| HDD with clean SMART that clicks or disappears | nothing | mechanical or electrical fault with no counter | do not power it on again; follow first aid guidance and contact a laboratory |
| SSD with clean SMART that suddenly becomes unreadable | nothing | controller/firmware failure (HPE at 32,768 h, SanDisk Extreme V2) | power it off, do not attempt repairs, contact a laboratory |
| Drive in a USB enclosure without SMART data | blank or „unknown" | the bridge does not pass through the commands | use -d sat or connect the drive directly to SATA |
What PASSED Means and Why a Short Self-Test Proves Nothing
For an ATA drive, smartctl -H displays PASSED when the SMART RETURN STATUS command does not report failure and no attribute marked as pre-fail has fallen to or below its threshold. The manufacturer sets the thresholds, and the tool compares normalized values rather than raw counts: a drive with 12 pending sectors and 40 reallocated sectors remains PASSED as long as its normalized value stays above the threshold. On NVMe, the test is even narrower—PASSED only means that the Critical Warning byte is zero.
CrystalDiskInfo calculates health differently. It displays „Caution„ as soon as the raw value of attributes 05, C5, or C6 reaches the configured threshold, which is 1 by default, or when an SSD has 10% or less remaining life. It displays „Bad" only when a value falls below the manufacturer’s threshold. Two tools, two definitions of health—and neither reads the entire surface.
A short self-test checks the electronics, mechanics, and only part of the drive’s readable area. According to Wikipedia, it usually takes less than two minutes; according to the smartctl manual, it „usually takes under ten minutes." An extended test reads the entire surface and takes hours. A successful short test tells you nothing about a sector near the end of a platter that the test never reached.
How to Read the Data in CrystalDiskInfo and smartctl
smartctl -a /dev/sda # health, attributes, self-test log
smartctl -x /dev/sda # also device statistics, SCT temperatures, SATA phy log
smartctl -a -d sat /dev/sdb # drive in a USB enclosure via SAT
smartctl -i /dev/nvme0 # model and firmware version (990 Pro: 0B2QJXD7 vs. 1B2QJXD7)
smartctl -a /dev/nvme0 # NVMe: Critical Warning, spare, Percentage Used, Media Errors
nvme smart-log /dev/nvme0 # the same information from nvme-cliThe -x output is useful in disputed cases: the SATA phy log shows whether attribute 199 errors increase with a particular cable, while on a WD Green you can see attribute 193 alongside 5 and 197 in one table. In CrystalDiskInfo, look at the raw value and the firmware version in the header. The tool takes attribute names from a generic table, so on an MX500 you will see „Current Pending Sector Count" even where Micron means ECC.
When to Back Up Immediately and When to Contact a Laboratory
Back up immediately while the drive is still readable: attributes 5, 197, or 198 are non-zero and higher on the second reading; 187 is greater than 0; or, on NVMe, Critical Warning is non-zero, Available Spare is below its threshold, or Media Errors are rising. Copy in a single pass, sector by sector, and read-only—we explain why you should not copy file by file or make repeated attempts in the article about bad sectors.
Go straight to a laboratory: the drive clicks or grinds, is not visible in the BIOS, reports zero capacity, disconnects under load, its SMART data cannot be read, or pending sectors increase within hours. The same applies to members of a RAID array and to an SSD that has suddenly stopped responding—every additional power-on attempt risks the only copy of your data.
Do nothing: an MX500 with attribute 197 flickering and attribute 5 at zero; a 990 Pro with a low percentage left by the old bug but no Media Errors; or a WD Green with a high 193 and zeros in 5/197/198. In these cases, adjust the timer or update the tool’s database instead of buying a new drive.
What Not to Do
- Run a long self-test on a drive with pending sectors. Hours of reading across weakened areas are exactly the kind of stress that can turn pending sectors into uncorrectable ones.
- Run chkdsk /r or a sector „repair." It writes to the drive and forces remapping; we cover when chkdsk helps and when it does not separately.
- Restart and reconnect the drive repeatedly. Every spin-up and every USB connection sends the heads over the damaged area again.
- Flash firmware on the only copy of your data. Even a supported update can fail; back up first.
- Silence a warning by increasing the threshold in CrystalDiskInfo. Changing the display does not repair the drive. It only makes sense when you know why the attribute is misleading, as with the MX500 or IntelliPark.
- Make a decision based on one number. Discarding a healthy 990 Pro because of its percentage is the same mistake as leaving a clicking drive in service because it says PASSED.
Frequently Asked Questions
My 990 Pro Shows 78% Health After a Few Months. Is It Faulty?
Check the firmware version first. If the drive ran on 0B2QJXD7, the percentage is a remnant of the bug fixed by firmware 1B2QJXD7 in February 2023; the value does not reset after the update. Media and Data Integrity Errors, Critical Warning, and Available Spare are what matter. If those are normal, the drive is operating normally.
CrystalDiskInfo Shows „Caution" for My MX500 Because of C5. Should I Replace It?
Not because of this. Micron uses attribute 197 to count ECC events, not sectors awaiting remapping, and the value fluctuates between 0 and 1. Monitor attributes 5 and 202. In smartmontools, update the drive database so that the attribute is given the correct name.
My WD Green Has a Load Cycle Count of 900,000. Is It Dying?
The number means that the head-parking mechanism has exceeded the figure it was designed for—300,000 cycles—not that the drive is failing. If attributes 5, 197, and 198 are zero, adjust the idle3 timer and fully power the drive off and back on. However, I would not continue using such a drive in a NAS or server without a backup.
SMART Says PASSED, but the Drive Is Clicking. Which Should I Trust?
The sound. PASSED only means that no pre-fail attribute has reached its threshold; SMART does not count a mechanical fault in the head or motor. According to Backblaze, almost a quarter of failures occurred with zeros in all five key attributes.
Can I Trust an SSD’s Health Percentage?
As the manufacturer’s estimate, yes; as a diagnosis, no. According to the NVMe specification, Percentage Used is a manufacturer estimate. A value of 100 does not necessarily mean failure, and it can rise above 100. An SSD that fails because of its controller or firmware may show a healthy percentage until the very last second.
If you are unsure which category your drive belongs to—its numbers look alarming but it works normally, or SMART is silent while the drive behaves strangely—leave it powered off and contact us. We perform diagnostics in our own laboratory in Brno, access drives only in read-only mode, and first tell you whether it is a false alarm, an immediate-backup case, or a job for data recovery.
Sources
- Tom’s Hardware: Samsung 990 Pro Firmware Update Addresses Failing SSD Health (13 February 2023)
- Rossmann Group: Samsung 990 Pro Firmware Bug and Health Loss
- Micron TN-FD-22: Client SATA SSD SMART Attribute Reference, Rev. E (PDF copy)
- smartmontools: drivedb.h — Crucial/Micron Client SSDs entry (attribute 197 Current_Pending_ECC_Cnt)
- Proxmox forum: smartd false positive SSD CurrentPendingSector? (MX500)
- Western Digital: WD Green Desktop Hard Drives — Spec Sheet 2879-771438 (300,000 load/unload cycles)
- StorageReview: How to Stop Excessive Load Cycles on the WD Caviar Green WD20EARS with WDIDLE3
- Manjaro Wiki: Western Digital Green — Drive Fix — Linux
- idle3-tools (SourceForge)
- WD Community: New WD30EFRX Red Drive Idle3 Timer Set to 8 Seconds (High LCC in NAS)
- Backblaze: What SMART Stats Tell Us About Hard Drives (2016)
- Backblaze: Hard Drive SMART Stats (2014)
- Computerworld: The 5 SMART stats that actually predict hard drive failure
- Wikipedia: Self-Monitoring, Analysis and Reporting Technology
- smartctl(8) — manual page (Debian)
- libnvme: nvme_smart_log — SMART / Health Information Log (field definitions)
- Crystal Dew World: CrystalDiskInfo — Health Status
- Crystal Dew World: CrystalDiskInfo — Health Status Setting
- BleepingComputer: HP Warns That Some SSD Drives Will Fail at 32,768 Hours of Use (2019)
- Tom’s Hardware: WD’s SSD Failures Stoke Class Action Lawsuit Over SanDisk Extreme Pro (2023)
Where to find us
ITHOPE is a company based in Brno, Czechia: we do data recovery from disks, RAID and NAS, PC and laptop service and IT for business, all in our own lab at Viniční 3587/216, 615 00 Brno.
Phone +420 775 556 063 (24/7 line). You can bring the device in person, or send it by courier (Zásilkovna, PPL).