How to check a used server before you pay for it
Most bad used-server deals announce themselves in advance, if you know where to look. Five checks, the paperwork, the management logs, the drives, the burn-in, the pre-shipment evidence, in the order you should run them.
Paper first: the serial tells the story
Before anyone plugs anything in, check the identity of the machine. Every enterprise server carries a vendor-registered identifier. Dell calls it a service tag, HPE and Lenovo use serial numbers, and the vendor's public lookup returns the configuration as it left the factory, the ship date and the warranty history. Five minutes of typing tells you whether the "2022 machine" was built in 2019, and whether the CPUs and memory in the listing match what the factory installed.
One result ends the conversation immediately: a serial that does not match the chassis, or does not resolve at the vendor at all. That is either a franken-build assembled from parts or something with a history nobody wants to discuss. Walk away, there is no discount that prices in an unverifiable machine.
Read the management controller logs
The baseboard management controller keeps its own record, and it is harder to fake than a listing. iDRAC on Dell, iLO on HPE, XCC on Lenovo, each logs what the seller cannot casually edit: total power-on hours, thermal events, corrected and uncorrected ECC memory errors, PSU faults, fan failures. Ask for a log export or a screen-share into the controller before you commit. A seller who refuses is hiding the history.
What healthy looks like: power-on hours consistent with the machine's claimed age and story (a "lightly used" box with 40,000 hours ran hot for four and a half years straight), no repeating thermal excursions, a corrected-ECC count of zero or nearly so, and no PSU failure entries. A single old event is normal; repeating ones are not.
Drives are the wear item
If anything in a used server is going to fail in your first year, it is a drive. SMART data is the tool here: reallocated sector count and pending sectors are the direct signal of a disk on its way out, and power-on hours tell you how much life is already spent. A chassis full of five-year-old drives with six-figure hour counts can be honest hardware, but you should price it as chassis plus drive replacements, not as a working system.
This is why our own quotes usually pair a refurbished chassis with new or low-hour drives. Depreciation sits in the chassis; risk sits in the drives. Splitting the two is the whole trick of buying used well.
What a real burn-in runs
A machine that posts to BIOS has proven almost nothing. Our standard burn-in runs 24–72 hours depending on configuration, and it is built to force marginal components to fail on our bench instead of in your rack. The test cycle covers:
- Full memory test passes, hours of it, because marginal DIMMs pass short tests.
- Sustained CPU load, and GPU stress tests on accelerated configs, checking thermals under load, not just function.
- Drive surface scans and SMART before/after comparison.
- Network throughput on every port at line rate.
Before the money moves
The last mile of verification is evidence of your machine, not a machine. Before the balance is due you get photos and video of the exact unit that will ship, serial numbers visible and matching the invoice, plus the burn-in report for that serial: what was run, for how long, temperatures, and a pass/fail per component with anything replaced noted.
Staged payment ties it together. A deposit starts the work; the balance is due only after you have reviewed the evidence. That ordering, proof first, money second, is the best protection in a cross-border hardware deal, and any seller who resists it has explained why.
Sources & further reading
The checks above lean on two things anyone can consult independently, the drive-health standard and vendor identity lookups:
- S.M.A.R.T. (Self-Monitoring, Analysis and Reporting Technology), the standard behind reallocated- and pending-sector drive checks: https://en.wikipedia.org/wiki/S.M.A.R.T.
- Backblaze Drive Stats, public failure-rate data, useful for judging how much life a given drive model has left: https://www.backblaze.com/cloud-storage/resources/hard-drive-test-data
- Vendor identity lookups. Dell decodes a service tag at dell.com/support, and HPE and Lenovo resolve serial numbers through their own support portals (search each vendor's support site directly, as the lookup pages sit behind bot protection).
Frequently asked questions
What happens if a server fails burn-in?
The failed component is swapped and the test cycle re-runs, before shipment, at our cost. That is the point of doing burn-in on our bench: failures are our problem there and yours after. If a swap moves the ship date, we tell you rather than quietly shortening the test.
Can I see the iDRAC / iLO logs before paying the balance?
Yes, ask and we include a log export or screenshots with the pre-shipment evidence: power-on hours, ECC counts, thermal history and PSU events for your specific unit. For larger orders we can arrange a live remote session into the controller.
Do you share the actual test reports?
Yes, per unit and tied to the serial number: tests run, durations, temperatures, results, and any parts replaced during burn-in. It arrives with the pre-shipment photos, before your balance is due.
Related
Specs and ships enterprise and GPU servers from our Hangzhou warehouse. These guides come from real orders and burn-in benches, not spec sheets.
About QingzhiGet a quote in 1 hour
Send us your configuration or your requirement — we confirm parts, price and lead time, and help you make sense of the import side.