Storage · 2026-10-02
Test a new drive before you trust it
The exact commands to run on a brand-new SSD or hard drive: SMART baseline, capacity verification, a full-surface write and read, and how to tell a used drive from a new one.
Half an hour of testing on an empty drive is cheaper than discovering the problem with your only copy of a 90 GB model on it. Do this before anything is written that you care about, and on a machine with nothing else on it — a fake drive is also an untrusted device.
Everything below is destructive. It erases the drive. That is the point: a full-surface write and read-back is the only test that finds fake capacity, bad blocks and intermittent errors.
1. Read the usage history before writing anything
sudo smartctl -a /dev/sdX
On a genuinely new drive, these should be at or near zero:
| ID | Attribute | What a new drive shows |
|---|---|---|
| 9 | Power_On_Hours | near zero |
| 12 | Power_Cycle_Count | single digits |
| 5 | Reallocated_Sector_Ct | 0 |
| 197 | Current_Pending_Sector | 0 |
| 199 | UDMA_CRC_Error_Count | 0 |
| 241 | Total_LBAs_Written | near zero |
| 194 | Temperature | within the rated range |
Non-zero hours, cycles or LBAs written mean the drive has been used. Note the values down as your baseline: you will want them again after the burn-in, to compare.
On a Seagate drive, also read the Field Accessible Reliability Metrics:
sudo smartctl -l farm /dev/sdX
FARM records hours per head and survives some counter resets, so a large gap between SMART power-on hours and FARM power-on hours means the SMART counters were wiped. It has been defeated before, so treat agreement as a green flag, not proof.
For an NVMe drive, the equivalent health read is:
sudo nvme smart-log /dev/nvme0
Percentage Used is a vendor estimate and can pass 100%. Data Units Written counts host writes, not raw NAND writes. Watch Available Spare, Critical Warning, media errors and temperature.
2. Prove the capacity is real
sudo f3probe --destructive --time-ops /dev/sdX
f3probe measures the real size against the advertised size and classifies the device; the most common fake verdict is “limbo”. Then fill and verify the whole thing:
sudo f3write /mnt/test && sudo f3read /mnt/test
f3write fills the file system with 1 GB files and f3read validates every one of them. The summary reports Data OK against Data LOST: any lost data means the drive is lying about its size. On Windows, h2testw does the same job.
3. Write the entire surface and read it back
sudo badblocks -wsv -b 4096 /dev/sdX
badblocks writes test patterns (0xaa, 0x55, 0xff, 0x00) to every block and reads each one back. Expect zero read errors, zero write errors, zero corruptions. One bad block on a new drive is a return, not something to work around.
Watch the temperature while it runs (sudo smartctl -a /dev/sdX in another terminal, attribute 194). Sustained writes are the only situation that reveals thermal throttling, and a drive that cooks in an M.2 slot next to a GPU will do it on your checkpoint writes too.
4. Run the drive’s own self-test
sudo smartctl -t long /dev/sdX
sudo smartctl -l selftest /dev/sdX
sudo smartctl -l error /dev/sdX
The extended self-test and the device error log catch controller-level faults the surface test misses.
5. Keep the baseline
Record the SMART output, the purchase date, the serial and the seller. When the drive is two years old, that file is the only way to know whether it is degrading or was always like this. smartd will warn you about rising reallocated sectors, spare-area loss and temperature without you checking by hand.
For NVMe drives, also confirm the link negotiated what you paid for:
sudo lspci -vv | grep -A3 -i nvme # link speed and width
sudo nvme id-ns /dev/nvme0 -H # namespace size, format
A PCIe 4.0 x4 drive that trains at x2 or at 3.0 has silently lost half its bandwidth, and slots, risers and bifurcated boards do this more often than people expect.
Used, or just new-looking?
The packaging is not evidence. These are:
- SMART power-on hours and cycles, plus the FARM comparison above
- per-head operating hours inside FARM, which survive a global counter reset
- a manufacturing date more than about six months older than your purchase date
- a missing original serial/barcode sticker, or a serial that fails the maker’s warranty lookup
- scuffs on the chassis of a drive advertised as sealed
- a firmware version that does not follow the vendor’s naming convention
- a negotiated interface generation lower than the model specifies
- a third-party controller marking and a missing DRAM chip where the retail model has one
What warranty actually covers
- Serial lookup decides everything. Seagate and Western Digital both verify by serial number, and WD states that the warranty term depends on the country in which the product was first purchased (Seagate, WD).
- OEM and bulk drives do not carry the retail warranty. The fraudulent “new” Exos drives were OEM models, so Seagate’s five-year term did not apply to them (Tom’s Hardware).
- Buying across regions moves the claim. If the drive came from an authorised distributor outside your region, you claim through that reseller, not locally.
- “Renewed” is a seller guarantee. On Amazon Renewed that is 90 days for Acceptable/Good/Excellent and 365 days for Premium — not a manufacturer warranty (Amazon).
- A single passing test proves nothing. The counterfeits that fooled the capacity check also fooled the benchmarks, and the FARM log has been altered in the wild. Combine checks, keep the receipt, and buy where the channel is contractually bound to authorised distribution.
Background on how the fraud works, and where it comes from, is in How drives get faked.