Enterprise SSD Percentage Used hit 100%: is it failing, and do you replace it?

1. 1.Print the SMART/Health Information log and read the Critical Warning byte first, bit by bit. Percentage Used is a vendor-specific estimate of life used, based on actual usage and the maker's prediction of NAND life, updated once per power-on hour. At 100 the estimated endurance is consumed, which the NVMe definition says may not indicate a failure.

   nvme smart-log /dev/nvme0

   Readingcritical_warning 0: the drive flags nothing. Nonzero: 0x01 available spare below threshold, 0x02 temperature out of range, 0x04 NVM subsystem reliability degraded, 0x08 media placed in read-only mode, 0x10 volatile memory backup failed. The bits describe the current state and are not persistent. Compare avail_spare (a normalised 0-100% of remaining spare) with spare_thresh; some nvme-cli builds spell them 'Available Spare' and 'Available Spare Threshold'. The threshold is set per drive, and Intel states 10% for its Optane products only. media_errors counts unrecovered data integrity errors. percentage_used can pass 100, and 255 is the field's ceiling: anything above 254 reads 255.
2. 2.Cross-check with smartctl, which names the bit that is set.

   smartctl -x /dev/nvme0

   Reading'SMART overall-health self-assessment test result: FAILED' appears for any nonzero Critical Warning bit, followed by the bit's meaning, for example '- NVM subsystem reliability has been degraded'. smartctl does not use Percentage Used to decide the result, so FAILED at 100% means the firmware set a bit. In the forum reports behind this page, drives past 100% showed Critical Warning 0x04 with Available Spare at 100%: that is what those drives did, not a vendor rule. A FAILED with bit 0x01 or 0x08 is a different and more serious state.
3. 3.Check what a script or monitor keyed on smartctl's exit code actually sees.

   smartctl -H /dev/nvme0; echo $?

   ReadingThe exit status carries the failure bit only when Critical Warning is nonzero, so anything keyed on smartctl's exit code alarms on the warning bit, not on Percentage Used.
4. 4.Optional: on drives that implement the Endurance Group Information log, read the same fields per endurance group. If the command errors, the drive does not provide the log; use the smart-log output above.

   nvme endurance-log /dev/nvme0 --group-id=1

   ReadingFields include critical_warning, avl_spare, avl_spare_threshold, percent_used, endurance_estimate and data_units_written. We have no source for the unit of endurance_estimate, so do not convert it.

1. 1.smartd's -H directive on NVMe checks the Critical Warning byte and logs a LOG_CRIT syslog message whenever any bit is set. How often it mails you is a separate setting, the -M directive: a warning mail goes out once by default, and daily by default when smartd runs with state persistence (its -s command-line option, not the smartd.conf -s self-test directive). The forum thread behind this page reports a daily 'disk failed' mail from a Proxmox host.
2. 2.Before you change the alert, check which smartmontools release you run, because what -H can ignore depends on it.

   smartd -V

   Reading7.4 or earlier: -H checks the whole Critical Warning byte with no option to ignore one bit, so removing -H for that device also silences bit 0x01 (spare below threshold) and bit 0x08 (read-only), the two replace signals; if you remove it, keep a separate check on Available Spare against its threshold and on those two bits. 7.5 (released 2025-04-30) or later: -H accepts a hexadecimal MASK that limits the check to the bits set in it, marked in the man page as an experimental feature. '-H 0xfb' (0xff without 0x04) ignores only the reliability-degraded bit and still alarms on 0x01 and 0x08; '-H 0xff' is the same as plain '-H'.
3. 3.If the monitor alarms on a percentage rather than on the warning byte, know which figure it reads. smartctl exports NVMe Percentage Used and the SAS percentage-used endurance indicator to the same JSON key, endurance_used.current_percent, so a threshold on that key fires on both interfaces alike.
4. 4.Read the error log, which smartd also watches.

   nvme error-log /dev/nvme0 -e 16

   ReadingEntries with status invalid command or invalid field in command come from the host side; smartd logs those at LOG_INFO and escalates to LOG_CRIT only for a new entry that is still present and device-related. Device-related entries together with a rising media_errors count are the concern.

1. 1.On a SAS SSD, read the endurance figure from SCSI log page 0x11 (Solid State Media).

   smartctl -l ssd /dev/sdX

   Reading'Percentage used endurance indicator: N%': 0 is new, 100 is the end of the manufacturer-projected lifetime, and the value may reach 255. The overall 'SMART Health Status' comes from the Informational Exceptions log, not from this percentage.
2. 2.If the SAS drive sits behind a RAID controller or you reach it through an SG device, address it as SCSI so smartctl does not send it ATA commands.

   smartctl -a -d scsi /dev/sgN

   ReadingThe same endurance indicator and health status as above, read through the SG device.
3. 3.On a SATA SSD, print the ATA Device Statistics log.

   smartctl -l devstat /dev/sdX

   ReadingPage 7, 'Solid State Device Statistics', holds 'Percentage Used Endurance Indicator' on drives that support the Device Statistics log; not every drive does, and smartctl -l ssd on ATA does the same thing. A vendor's own wear figure is separate and vendor-specific: Solidigm's warranty names its SATA figure 'Media Wear Out Indicator', as reported by the Solidigm Storage Tool, and it falls toward 1 rather than rising toward 100.

1. 1.Read the drive's endurance on the iDRAC physical disk page. iDRAC reports Remaining Rated Write Endurance, which counts down from 100% when new: the opposite direction to NVMe Percentage Used and the SCSI and ATA indicators. Oracle's ZFS Storage Appliance also counts down.

   iDRAC web UI: the physical disk page for the drive (Remaining Rated Write Endurance)

   ReadingOn certain Dell enterprise SAS and SATA SSDs, a reading below a threshold raises a SMART predictive failure with the message 'SSD Write Endurance is less than the threshold value of <Remaining Write Rated Endurance>'. Dell's article states no threshold percentage.
2. 2.Collect a TSR from the iDRAC first. Dell's steps for this event start there, and the step after this one deletes log entries the TSR would otherwise have captured.
3. 3.Clearing the System Event Log permanently deletes its entries, so do this only once the TSR is saved. Dell's steps then reset the iDRAC. Dell scopes this article to VxRail.
   Destructive. This command destroys what is on the device and cannot be undone. Confirm the device path twice and have a backup of anything you still need before running it.

   iDRAC web UI: clear the System Event Log (SEL), then reset iDRAC

4. 4.Wait 5-10 minutes and re-check Remaining Rated Write Endurance on the physical disk page. If the low reading persists, Dell's step is to contact support with the TSR. Read the drive from the OS as well, with the NVMe or SAS/SATA paths above.
   ReadingA ceph-users mailing list report describes iDRAC9 6.10 showing 0% write endurance on every non-Dell-certified SSD while OMSA showed the real figure. That is a forum report, not a Dell statement, but it is a reason to read the drive itself before acting on iDRAC alone.

1. 1.Identify the model and firmware revision, then find the drive's rated TBW or DWPD and warranty years in the vendor's data sheet.

   nvme id-ctrl -H /dev/nvme0

   ReadingTBW is the JEDEC JESD218 endurance rating: terabytes written by the host. DWPD is full-capacity writes per day under a specified workload over a specified lifetime, so rated total writes = capacity x DWPD x 365 x rated years.
2. 2.Record Data Units Written now and again after N days of normal load. One unit is 1,000 units of 512 bytes of host data, excluding metadata.

   nvme smart-log /dev/nvme0 -o json

   ReadingTB written so far = data_units_written x 512,000 / 10^12. Write rate per day = (second reading - first reading) x 512,000 / 10^12 / N. Remaining rated writes = rated TBW (or capacity x DWPD x 365 x rated years) - TB written so far. Days left on the rating = remaining rated writes / write rate per day. If TB written already exceeds the rating, the rating is spent and the Critical Warning bits are what describe the drive's state.
3. 3.Make the call. Replace now if Critical Warning bit 0x01 or 0x08 is set, Available Spare is below its threshold, or media_errors keeps rising. Plan a replacement if the days left on the rating are fewer than the drive's remaining planned service. Otherwise the 100% alarm on its own is an estimate and warranty event, not a replace signal.
4. 4.Check the warranty before planning on an RMA. Solidigm's data center warranty terminates immediately when Percentage Used reaches or exceeds 100 on a PCIe drive, or when the Media Wear Out Indicator reaches 1 on a SATA drive, as reported by the Solidigm Storage Tool. Dell's limited hardware warranty does not cover a device that has reached its maximum device life.

Percentage Used is a vendor-specific estimate based on actual usage and the maker's prediction of NAND life, not a direct count of writes. In one Proxmox/Ceph thread, two of three identical NVMe drives with about 20.5TB written each read 150 and 255 (the field's ceiling, meaning above 254) while the third read 6; no resolution was posted, and we have no vendor document that explains such a gap.

We found no vendor document that states what a drive does after Percentage Used passes 100, whether it keeps working, slows or goes read-only. A drive that does enter read-only mode reports it through Critical Warning bit 0x08, so that bit is the one to watch.

Under Solidigm's data center warranty, no: it covers only the original purchaser who bought the drive in sealed packaging from Solidigm or an authorized distributor, for as long as they own it. We read only Solidigm's terms on this point, so check each maker's own warranty.

Samsung's SSD Limited Warranty ends at the shorter of the warranty period or the date the SSD exceeds its TBW threshold, with the threshold in each product's data sheet; the TBW table in that document lists retail models, and we did not read a Samsung data center warranty. Kioxia lists 5 years for its enterprise and data center SSDs, and we found no endurance clause on Kioxia's EMEA warranty page.