A renewed 24TB Ultrastar DC HC580 sells for about $819 on Amazon this week, which works out to $34.13 per terabyte, while a new 22TB HC580 from the same family costs about $1,252 from a third-party seller, or $56.91 per terabyte. That 40 percent gap per terabyte is why recertified hard drives fill so many homelab racks, and the gap is also why used mining drives with reset hour counters were resold as new in early 2025.
This guide answers whether recertified hard drives are worth the saving, explains what separates manufacturer recertified, refurbished, Amazon Renewed and plain used drives, and then walks through the Linux test workflow that decides whether a drive stays: a SMART baseline with smartctl, the Seagate FARM log that exposed the fraud, short and extended self tests, a destructive badblocks burn in sized correctly for drives above 16TB, and smartd monitoring for the return window.
Tested October 2026 with smartmontools 7.5 (Ubuntu 26.04).
Are recertified hard drives worth it?
Yes for the second or third copy of data, and for parity arrays where one failed drive is an inconvenience rather than a loss. No for the only copy of anything, because the warranty on most listings is the reseller cover, not the manufacturer warranty, and on Amazon the term ranges from 90 days to five years on drives of the same class.
The saving is real and large: every recertified drive priced for this guide landed between $29.39 and $34.45 per terabyte, against roughly $50 to $57 per terabyte for new Ultrastar HC580 and HC590 drives. The risk is also real, but it is testable. Every surface defect that matters shows up in the six steps below, and a drive that fails them goes back inside the return window.
Recertified vs refurbished vs renewed vs used
The four labels are used loosely on marketplaces, so the label on a listing says less than the warranty line next to it. The table separates them by who tested the drive and who pays when it dies.
| Label | Who tested it | Warranty source | Typical term |
|---|---|---|---|
| Manufacturer recertified (Seagate direct) | Seagate: sanitize, test, certify | Seagate | 6 months |
| Manufacturer recertified (WD store) | Western Digital | Western Digital | 1 year |
| Manufacturer recertified via reseller | The manufacturer, resold by ServerPartDeals, goHardDrive and others | The reseller | 90 days to 5 years, per listing |
| Refurbished | A third party or the reseller in house | The reseller | Varies |
| Amazon Renewed | Depends on the seller behind the listing | Seller, plus the Amazon Renewed Guarantee on eligible listings | 90 days (1 year for Renewed Premium) |
| Used or pulled | Nobody, sold as removed from a server | Seller only, if any | Often none |
Seagate describes its own program as a three step process of sanitize, test and certify, with a six month warranty, and it states that prime drive specifications and warranties do not apply to recertified products (Seagate recertified program). Western Digital sells recertified drives on its own store with a one year limited warranty. ServerPartDeals defines manufacturer recertified as tested and approved by the original manufacturer without a manufacturer warranty, so the reseller handles every claim.
Recertified drive prices compared with new
Prices below came from the Amazon Creators API on 6 October 2026 for US listings, so treat each one as approximate and check the live price before ordering. The per terabyte column uses the decimal capacity printed on the label.
| Drive | Condition | Price (approx.) | Per TB | Listing warranty |
|---|---|---|---|---|
| Seagate Exos 28TB ST28000NM000C | Renewed | about $920 | $32.86 | 90 days |
| WD Ultrastar DC HC550 18TB | Renewed | about $529 | $29.39 | 90 days |
| WD Ultrastar DC HC580 24TB | Renewed | about $819 | $34.13 | 2 years, seller |
| Seagate Exos 24TB ST24000NM000C | Renewed | about $790 | $32.92 | Amazon Renewed Guarantee |
| WD Ultrastar DC HC560 20TB | Renewed | about $689 | $34.45 | 2 years, seller |
| WD Ultrastar DC HC580 22TB | New | about $1,252 | $56.91 | Manufacturer, if the serial is a retail unit |
| WD Ultrastar DC HC590 26TB | New | about $1,313 | $50.50 | Manufacturer, if the serial is a retail unit |
Two patterns stand out. Four of the five renewed drives sit within $1.60 per terabyte of each other, and the warranty column varies more than the price column, so a listing with a two year seller warranty at $34.13 per terabyte is a better buy than a 90 day listing at $32.86. The wider price picture, including why new drives cost what they do this year, is covered in why RAM and storage prices increased.
What the reliability data says
No public failure dataset exists for recertified drives. Backblaze, ServerPartDeals and the manufacturers publish nothing on annualized failure rates for second life units, so any claim that recertified drives fail at a specific rate is a guess. What does exist is strong data on how failure rate changes with age, and that data argues for recertified drives more than against them.
The 2025 Backblaze bathtub curve analysis, covering 317,230 drives, puts the failure peak at 4.25 percent at ten years and three months, with drives under one year old at about 1.30 percent and a flat failure rate through most of the working life (Backblaze bathtub curve 2025). A recertified drive with two or three years of service sits in the flat middle of that curve, past infant mortality and years away from the wear out spike. The infant mortality phase is exactly what the burn in below is designed to surface before data lands on the drive.
Failure rates vary widely by model. In the Backblaze Q2 2026 report the Exos X16 16TB ran at a 0.68 percent lifetime rate across 35,100 drives, while the Exos 24TB ST24000NM002H sat at 3.24 percent across 9,631 drives. Check the Backblaze row for the exact model before buying any drive, new or recertified.
Why SMART hours on a recertified drive can lie
In January 2025 heise reported Seagate Exos drives sold as new in Germany that showed clean SMART data but 10,000 to 22,000 power on hours in the Seagate FARM log (heise investigation). Within weeks more than 200 cases surfaced across Europe, Australia, Thailand and Japan, with TechSpot counting 15,000 to 50,000 prior hours on affected drives, and later reporting pointed to Chia mining farms in China as the likely source. Seagate confirmed that SMART values can be reset and that the FARM figures were probably correct.
Two consequences follow for anyone buying recertified. First, a power on hours value of zero proves nothing, because ServerPartDeals states that manufacturer recertification resets the counter to zero by design. Second, by March 2025 the fraudsters were also clearing the FARM power on hours on 20TB Exos drives, so heise recommends comparing the per head operating hours inside FARM, which are harder to fake, along with the production date and the label. FARM exists only on Seagate drives, which leaves WD and Toshiba buyers with SMART alone.
Recertified drives worth buying right now
Every pick below is a CMR, helium sealed enterprise drive confirmed against the manufacturer datasheet, and each one names the compromise that matters before ordering. For NAS drives with full manufacturer warranties, see the best NAS hard drives guide.
WD Ultrastar DC HC580 24TB: best warranty per dollar
Buy this one when the listing warranty matters more to you than the last dollar per terabyte.
The renewed HC580 listing from ServerPartDeals carries a two year seller warranty at about $819, which ties the HC560 20TB listing for the longest cover on any recertified drive priced for this guide. The datasheet lists a 10 disk CMR design with OptiNAND, sustained transfer up to 298 MB/s and a 550TB per year workload rating. Skip it if the enclosure uses an older power supply or backplane: the WUH722424ALE604 part on this listing is a Power Disable model, and a drive that never spins up on a 3.3V pin 3 rail is a wiring problem, not a dead drive.
Seagate Exos 28TB ST28000NM000C: most capacity per bay
This is the drive for a chassis that is short on bays, since it packs the most terabytes into one slot.
The ST28000NM000C appears in the Seagate recertified Exos datasheet as a 28TB CMR helium drive with a 2.5 million hour MTBF rating, and the renewed listing at about $920 works out to $32.86 per terabyte, the lowest per terabyte of any drive above 18TB in the price table. The same datasheet gives every recertified Exos model one flat 190 MB/s sustained transfer rating, against 285 MB/s on the new Exos X20 datasheet. That is a conservative rating rather than a measurement, but planned at that rate a full badblocks pass takes about two weeks. Skip it if a 90 day listing warranty is too short: the Amazon listing states 90 days, while the same drive bought direct from ServerPartDeals carries a three year seller warranty, and this model is also the one where FARM checks matter most.
WD Ultrastar DC HC550 18TB: cheapest drive to buy
Start here if the total spend per drive is the limit rather than the cost per terabyte.
At about $529 the renewed HC550 is the cheapest drive in the price table, at $29.39 per terabyte, and the platform has a long field record: the 16TB HC550 ran at a 0.56 percent lifetime failure rate across 26,827 drives in the Backblaze Q2 2026 report. The listing describes a pin 3 Power Disable drive, the same wiring caveat as the HC580. Skip it if the drive will hold anything without a second copy, because the listing warranty is 90 days, which covers the burn in and very little after it.
One more listing trap: the HC560 20TB appears on Amazon both as WUH722020BLE601, which the renewed listing marks as Power Disable, and as WUH722020BLE6L4, which it does not. The two suffixes look alike in a listing title and behave differently in a consumer case. For SMR traps on the cheaper end of the market, the CMR vs SMR guide lists the models to avoid.
Step 1: Install smartmontools
The FARM log needs smartmontools 7.4 or newer, and 7.5 fixes a unit bug that printed per head hours in seconds. The version shipped by each distribution decides whether the fraud check in Step 3 works at all, and Proxmox VE 8 hosts are on the wrong side of that line.
| Distribution | smartmontools | smartctl -l farm |
|---|---|---|
| Ubuntu 26.04 LTS | 7.5 | Yes |
| Ubuntu 24.04 LTS | 7.4 | Yes, head hours shown in seconds |
| Debian 13 | 7.4 | Yes, head hours shown in seconds |
| Debian 12 and Proxmox VE 8 | 7.3 | No |
| Fedora 44 and 45 | 7.5 | Yes |
| Rocky Linux 10 | 7.4 | Yes, head hours shown in seconds |
| Rocky Linux 9 | 7.2 | No |
On Ubuntu and Debian:
sudo apt update
sudo apt install -y smartmontoolsOn Fedora, Rocky Linux and RHEL:
sudo dnf install -y smartmontoolsThe burn in in Step 5 runs inside tmux, which minimal server installs often lack. Add it with sudo apt install -y tmux or sudo dnf install -y tmux. Then confirm the smartmontools version before trusting any FARM output:
smartctl --version | head -1The Ubuntu 26.04 test box reported the release with the FARM fixes:
smartctl 7.5 2025-04-30 r5714 [x86_64-linux-7.0.0-15-generic] (local build)On a Proxmox VE 8 node, which reports 7.3, run the FARM check from a live Ubuntu 26.04 USB stick or another machine. Proxmox VE 9 is built on Debian 13, which ships 7.4. The node can still run every other step.
Step 2: Record the drive identity and SMART baseline
Find the device name by serial number first, because a destructive test on the wrong disk is the most expensive mistake in this whole workflow. Match the serial on the label to the serial printed here:
lsblk -d -o NAME,SIZE,ROTA,TRAN,MODEL,SERIALStore the serial in a variable, which keeps every file from here on named per drive when several drives are tested at once:
SERIAL=$(sudo smartctl -i /dev/sdX | awk -F: '/^Serial [Nn]umber/{gsub(/ /,"",$2); print $2}')
echo "$SERIAL"The awk pattern matches both the SATA spelling “Serial Number” and the SAS spelling “Serial number”. Then save the full SMART report, so the post burn in comparison has a reference:
sudo smartctl -x /dev/sdX | tee smart-before-${SERIAL}.txtThe lab has no spinning drives, so the baseline below comes from a SATA SSD on one of the Proxmox nodes. It makes a better teaching example than a fresh drive would, because it is a used disk that still reports a passing health check. These are the attributes from the real capture:
SMART overall-health self-assessment test result: PASSED
ID# ATTRIBUTE_NAME FLAGS VALUE WORST THRESH FAIL RAW_VALUE
5 Reallocated_Sector_Ct PO--CK 100 100 010 - 0
9 Power_On_Hours -O--CK 094 094 000 - 25256
12 Power_Cycle_Count -O--CK 099 099 000 - 742
187 Uncorrectable_Error_Cnt -O--CK 100 100 000 - 0
190 Airflow_Temperature_Cel -O--CK 050 030 000 - 50
199 CRC_Error_Count -OSRCK 093 093 000 - 6503The overall health line reads PASSED on a drive with 25,256 power on hours (2.9 years of continuous running) and 6,503 interface CRC errors. PASSED only means no normalized value has crossed its vendor threshold, and the thresholds are set so low that drives with pending sectors routinely pass. Read the raw values instead. The attributes that decide a used hard drive are these:
| ID | Attribute | Accept on arrival | Why it matters |
|---|---|---|---|
| 5 | Reallocated_Sector_Ct | 0 | Sectors already remapped; Backblaze investigates any value above zero |
| 187 | Reported_Uncorrect | 0 | Reads the drive could not correct |
| 188 | Command_Timeout | 0 or a small stable number | Operations that timed out, often power or cabling |
| 197 | Current_Pending_Sector | 0 | Unstable sectors waiting for a rewrite; one of the two best predictors in Backblaze data |
| 198 | Offline_Uncorrectable | 0 | Sectors that failed the offline scan |
| 199 | UDMA_CRC_Error_Count | Any, if it stops rising | Link errors between drive and controller, not media |
| 9 | Power_On_Hours | Compare with FARM on Seagate | Can be reset; zero is normal after manufacturer recertification |
| 193 | Load_Cycle_Count | Well under the datasheet rating | Head parks; the HC560 datasheet rates 600,000 |
Backblaze tracks attributes 5, 187, 188, 197 and 198, and in its fleet 76.7 percent of failed drives showed a raw value above zero on at least one of them, against 4.2 percent of working drives (Backblaze SMART stats). The 6,503 CRC errors on the lab SSD are counted on the SATA link between drive and controller rather than in the flash, so attribute 199 is judged on whether it keeps climbing after a cable or port change.
The self test log is a second history record. Each entry stores the drive lifetime hour at which the test ran, and on the lab SSD the newest entry jumps from 10,680 hours straight to 25,256, which shows nobody ran a test on that disk for about 14,600 hours:
Num Test_Description Status Remaining LifeTime(hours) LBA_of_first_error
# 1 Short offline Completed without error 00% 25256 -
# 2 Short offline Completed without error 00% 10680 -
# 3 Short offline Completed without error 00% 10657 -
# 4 Short offline Completed without error 00% 10633 -The newest entry is the short test run for this guide, and the long gap before it is the part worth reading on any used drive.
Step 3: Read the Seagate FARM log
On a Seagate drive, FARM (Field Access Reliability Metrics) is a vendor log kept separately from the standard SMART attributes, and it was the log that exposed the 2025 fraud. Print it and save it next to the SMART baseline:
sudo smartctl -l farm /dev/sdX | tee farm-${SERIAL}.txtCompare three things. The FARM power on hours against SMART attribute 9, the per head power on hours inside FARM against both, and the assembly date against the date on the label. A recertified drive is allowed to show zero SMART hours, but per head hours in the tens of thousands on a drive sold as new is a return, whatever the listing says.
On anything that is not a Seagate drive the command reports that the log is unsupported, which is what smartctl 7.5 printed for the virtual disk on the test box:
FARM log (SCSI Log page 0x3d, sub-page 0x3) not supportedWD and Toshiba drives have no FARM equivalent that smartctl reads, so for the two WD picks the SMART baseline, the self test log, the production date on the label and the seller warranty are the full set of evidence.
Step 4: Run the short and extended self tests
The short test takes about two minutes and catches gross faults before any time goes into the long test:
sudo smartctl -t short /dev/sdXsmartctl returns immediately and prints when the test will finish:
=== START OF OFFLINE IMMEDIATE AND SELF-TEST SECTION ===
Sending command: "Execute SMART Short self-test routine immediately in off-line mode".
Drive command "Execute SMART Short self-test routine immediately in off-line mode" successful.
Testing has begun.
Please wait 2 minutes for test to complete.
Test will complete after Mon Oct 5 17:51:12 2026 EAT
Use smartctl -X to abort test.Check progress, then the result once the time has passed:
sudo smartctl -c /dev/sdX | grep -A1 'Self-test execution'
sudo smartctl -l selftest /dev/sdXA finished test adds a new top entry that reads “Completed without error”. Then start the extended test, which reads the whole surface. The drive reports its own estimate in the “Extended self-test routine recommended polling time” line of smartctl -c, and on a 20TB class drive that figure runs past a day. A TrueNAS forum user measured an average of 1 day 12 hours 42 minutes on four 20TB WD Red Pro drives.
sudo smartctl -t long /dev/sdXKeep the drive on a direct SATA or HBA port for the whole run. If the log shows the extended test as interrupted or aborted, rerun it there, and treat a drive that still fails to complete it as a return.
Step 5: Burn in with badblocks
badblocks in write mode fills every block with four patterns (0xaa, 0x55, 0xff and 0x00) and reads each one back, so the drive sees eight full surface passes. It destroys all data on the target, and it refuses to run on a mounted file system, which the test box confirmed:
sudo badblocks -b 4096 -wsv /dev/sdXRun against the mounted scratch disk on the test box, the refusal came back before a single block was written:
/dev/sda is mounted; it's not safe to run badblocks!The bigger trap is the block size. badblocks stores block numbers as 32 bit values, so the last block number must stay under 4,294,967,296. With the default 1024 byte blocks that caps the drive at 4.4TB, and with 4096 byte blocks at 17.59TB. On a 20TB disk (a virtual disk of 20,000,588,955,648 bytes on the test box, just over 20TB) both the default and -b 4096 fail at once:
The exact error string, for anyone searching for it:
badblocks: Value too large for defined data type invalid end block (4882956288): must be 32-bit valueThe fix is -b 8192, which covers drives up to 35.18TB and therefore every capacity in this guide. Size the block to the drive:
| Drive capacity | Blocks at -b 4096 | Blocks at -b 8192 | Use |
|---|---|---|---|
| 16TB | 3,906,250,000 | 1,953,125,000 | -b 4096 |
| 18TB | 4,394,531,250 (too large) | 2,197,265,625 | -b 8192 |
| 20TB | 4,882,956,288 (too large) | 2,441,478,144 | -b 8192 |
| 24TB | 5,859,375,000 (too large) | 2,929,687,500 | -b 8192 |
| 28TB | 6,835,937,500 (too large) | 3,417,968,750 | -b 8192 |
The 20TB row uses the byte count measured on the test box; the other rows use the decimal label capacity, so the exact count on a real drive differs slightly. Check it with sudo blockdev --getsize64 /dev/sdX divided by the block size. Run the burn in inside tmux, because it outlives any SSH session, and write any bad block numbers to a file:
tmux new -s burnin
sudo badblocks -b 8192 -wsv -o badblocks-${SERIAL}.txt /dev/sdXA clean run on the 16 GiB scratch disk of the test box ended with the line every drive needs to print, and an empty output file. The timing of that run is not quoted here because the disk was a ZFS volume, and ZFS compresses the four fixed patterns, which makes the elapsed time meaningless for a hard drive:
Pass completed, 0 bad blocks found. (0/0/0 errors)Burn in time on a real drive is set by its sustained transfer rate. The full four pattern test equals eight passes over the surface, and the datasheet rate is the outer track maximum, so real runs on the WD drives take longer than these figures. The recertified Exos row uses the flat recertified rating, which real drives likely beat:
| Drive | Datasheet rate | Full -w (8 passes) | Single -t random (2 passes) |
|---|---|---|---|
| HC550 18TB | 269 MB/s | at least 6.2 days | at least 1.5 days |
| HC580 24TB | 298 MB/s | at least 7.5 days | at least 1.9 days |
| Exos 28TB recertified | 190 MB/s (recertified rating) | about 13.6 days | about 3.4 days |
Two weeks per drive eats a large share of a 90 day return window, so on the slower recertified Exos a single random pattern pass is the practical choice: sudo badblocks -b 8192 -wsv -t random -o badblocks-${SERIAL}.txt /dev/sdX. It writes and verifies every block once, which exercises the full surface and forces the drive to remap any weak sector, in a quarter of the time. Testing several drives in parallel works as long as each runs in its own tmux window and the power supply has headroom for every spindle at once.
Step 6: Compare SMART after the burn in
Capture the report again and compare the raw values that matter:
sudo smartctl -x /dev/sdX > smart-after-${SERIAL}.txt
grep -E '^ *(5|187|188|197|198|199) [A-Z]' smart-before-${SERIAL}.txt smart-after-${SERIAL}.txtThe decision rules are deliberately strict, because a return costs a shipping label while a failed drive in an array costs a resilver under load:
| After burn in | Verdict |
|---|---|
| badblocks output file not empty | Return |
| 5, 197 or 198 rose from the baseline | Return |
| 187 rose during the run | Return |
| 199 rose, everything else flat | Swap the cable or backplane slot, rerun the short test |
| Extended test aborted on a direct SATA port | Return |
| All raw values unchanged, extended test passed | Keep, then add to the pool |
A drive that passes goes into the array like any new disk. For pool layout on Proxmox, the Proxmox ZFS RAID levels guide covers the parity math, and recertified drives are a good fit for RAIDZ2, which survives two simultaneous drive failures.
What to watch in SMART during the return window
The burn in catches infant mortality, and smartd catches the slower failures that show up in the following weeks. A single DEVICESCAN line monitors every drive, runs a short test every night at 02:00 and a long test every Saturday at 03:00, and tracks temperature: a change of 4 degrees or a reading of 45 degrees Celsius is logged at info level, while 55 degrees is the critical limit that also triggers the mail. smartd ignores every line after the first DEVICESCAN, so comment out the distribution default before appending the new one. The new line keeps the two Debian defaults, -n standby so a sleeping drive is not woken for a check, and the smartd-runner mail hook. Run the pair once:
sudo sed -i 's/^DEVICESCAN -d removable/#&/' /etc/smartd.conf
echo 'DEVICESCAN -a -n standby -s (S/../.././02|L/../../6/03) -W 4,45,55 -m root -M exec /usr/share/smartmontools/smartd-runner' | sudo tee -a /etc/smartd.confThose paths are for Ubuntu and Debian. On Fedora, Rocky Linux and RHEL the file is /etc/smartmontools/smartd.conf: comment out its active line with sudo sed -i 's/^DEVICESCAN/#&/' on that path before appending, and leave the Debian only -M exec path off the new line. The -m root directive mails warnings to root, which only reaches a person if the host has a working mail relay. Validate the file with a single check pass before restarting the daemon:
sudo smartd -q onecheck -dOn the Proxmox node with the 850 EVO, a check pass with the same schedule and temperature directives printed this, including an info level note that the drive had already passed the 45 degree mark:
Device: /dev/sda [SAT], Samsung SSD 850 EVO 250GB, S/N:S2R5XXXXXXXXXXX, WWN:5-002538-XXXXXXXXX, FW:EMT02B6Q, 250 GB
Device: /dev/sda [SAT], is SMART capable. Adding to "monitor" list.
Monitoring 1 ATA/SATA, 0 SCSI/SAS and 1 NVMe devices
Device: /dev/sda [SAT], initial Temperature is 50 Celsius (Min/Max ??/50)
Device: /dev/sda [SAT], Temperature 50 Celsius reached limit of 45 Celsius (Min/Max ??/50)
Device: /dev/sda [SAT], previous self-test completed without error
Started with '-q onecheck' option. All devices successfully checked once.
smartd is exiting (exit status 0)Restart the daemon so it loads the new line. The unit is smartmontools on Ubuntu and Debian and smartd on Fedora and RHEL:
sudo systemctl restart smartmontools
sudo systemctl status smartmontools --no-pagerFor the rest of the return window, the number that matters is the trend in attributes 5, 197 and 198 rather than the absolute value. One reallocated sector that never moves again is common on old drives; a count that rises every week is a drive heading for failure, and on a 90 day listing it needs to go back before day 90, not after it. Recertified drives used as backup targets pair well with the enclosures in the external backup drives guide, which apply the same smartctl checks over USB.
