How to Test a Used Graphics Card Before Buying
Used GPUs are the best-value parts in PC building and the most confidently misdescribed. "Works perfectly" covers everything from a pampered card to one that artifacts the moment VRAM fills. The first three checks run in a browser in minutes; the rest use free desktop tools and turn a gamble into a measurement.
-
Inspect the card physically
Look for bent heatsink fins, cracked plastic, sag marks near the PCIe slot tab, and burn discolouration around the power connectors — melted or browned pins on a 12VHPWR/16-pin connector are disqualifying. Check the screws on the cooler shroud and backplate: chewed heads mean the card has been opened, which is fine if the seller says so (repaste) and suspicious if they claim it is untouched.
Needs a desktop tool or your own eyes — details in the step.
-
Verify the GPU is what the listing says
Plug the card into a PC and check the detected renderer string matches the model being sold. Fake cards — old GPUs re-flashed to report as newer models — are a real problem on marketplace sites, and they reveal themselves under load or in the renderer details. Cross-check the memory size and driver-reported model in the seller’s OS too (Task Manager or GPU-Z), not just the sticker on the cooler.
Open the GPU test → -
Run a live render and watch for artifacts
Flickering textures, stray coloured triangles, checkerboard patterns or a driver reset during even a light render point to failing VRAM or a dying core. A healthy card renders cleanly for minutes without a single visual glitch. Any artifact you can see in a browser test will be far worse in a game.
Run the WebGL render → -
Load the card and watch temperatures
Run a sustained stress test — FurMark, 3DMark stress test, or Heaven/Superposition looped for 15–20 minutes — with HWiNFO or MSI Afterburner logging. GPU core in the 60–80°C range under full load is normal; sustained readings pushing 90°C+, hot-spot deltas above ~25°C, or memory junction temperatures at their limit mean dried-out paste and pads at best, a cooling defect at worst.
Needs a desktop tool or your own eyes — details in the step.
-
Test the VRAM specifically
Video memory fails before cores do, and light tests barely touch it. Run a dedicated VRAM tester (memtest_vulkan or OCCT’s VRAM test) for a full pass. Memory errors that appear only when the VRAM is nearly full explain the classic "it crashes in some games but not others" pattern sellers describe as driver issues.
Needs a desktop tool or your own eyes — details in the step.
-
Listen to the fans
At idle, fans should be stopped or silent. Under load, listen for grinding, ticking or a rhythmic rattle — worn bearings. Spin each fan gently by hand with the power off: they should rotate freely and identically. Replacement fans exist for popular models but can be surprisingly hard to source for others; price a set before treating it as a small fault.
Needs a desktop tool or your own eyes — details in the step.
-
Test every display output
Plug a monitor into each DisplayPort and HDMI output in turn and confirm a clean image on all of them. A dead output is a common fault on cards that have been repaired or ridden hard, and the easiest thing for a seller to avoid demonstrating.
Open the screen test → -
Benchmark and compare against reference numbers
Run a standard benchmark (3DMark Time Spy, Superposition) and compare the score with published results for the same model — thousands are a search away. A card scoring 10–15 percent below typical for its model is throttling, power-limited or degraded. This single check catches most "it works fine" cards that do not, in fact, work fine.
Needs a desktop tool or your own eyes — details in the step.
-
Judge the mining/AI-farm question sensibly
You usually cannot prove a card’s history, so look for the correlates: a seller with many identical cards, mining BIOS or aggressive undervolt profiles, dust patterns from open-rack use, and connectors worn from re-plugging. Contrary to folklore, a mining card that ran cool and undervolted can be healthier than a gamer’s card that thermal-cycled daily — judge the card in front of you by the tests above, and price in the unknown.
Needs a desktop tool or your own eyes — details in the step.
Red flags — negotiate hard or walk away
- Any visual artifact at stock settings. There is no economic repair; it only gets worse.
- A renderer string, memory size or GPU-Z readout that does not match the model being sold.
- Browned, melted or heat-discoloured power connector pins.
- Benchmark scores well below published results for the same model after drivers are ruled out.
- VRAM test errors, or crashes that appear only when memory fills ("some games crash, drivers probably").
- A dead display output the seller "never uses".
- A seller who refuses a load test — "you can see it posts" is not a test.
- No receipt on a brand whose warranty needs one, priced as if warranted.
Frequently Asked Questions
How can I tell if a used graphics card is fake?
Trust what the card reports under load, not the sticker: the browser GPU test shows the renderer string, GPU-Z shows the die and memory configuration, and a benchmark shows whether performance matches the name. Re-flashed fakes fail at least one of the three.
Is it safe to buy a GPU that was used for mining?
Often yes, at the right price. Undervolted steady-state mining is gentler than daily thermal cycling; the wear lands on fans and paste, which are replaceable. Since history is unprovable, test thoroughly and price in a repaste.
What temperature is too hot for a used GPU?
60–80°C sustained under load is normal. Above 90°C sustained, a core-to-hot-spot gap over ~25°C, or memory junction at its limit means degraded paste and pads or a cooler fault — negotiate or walk.
What causes graphical artifacts on a GPU?
Failing VRAM, a degraded core, an unstable factory overclock, or cracked solder joints. At stock settings on a used card, artifacts are effectively terminal — do not buy, whatever the discount.
Does the manufacturer warranty transfer to a second owner?
Brand-dependent: some honour the serial number for any owner, others require the original invoice. Check the brand's policy and ask for the receipt — a transferable warranty is real added value.
Buying the whole PC, not just the card? Run the used laptop checklist for everything else, and system info to confirm the rest of the spec.