How to Test a Used GPU (graphics card) Before You Buy (20-Minute Check)
A used GPU (graphics card) can be a great deal — or someone else's problem. This is the exact 20-minute test to run before you hand over cash, with the real tools and the red flags that mean walk away.
The test kit
Cheap, Prime-fast tools that make this test reliable. (affiliate)
- USB-C / PCIe power meter (inline wattage) →confirms the card actually pulls its rated power and the PSU holds up under load
- Compressed-air duster →blow out dust before testing so thermals reflect the card, not a clogged heatsink
- Spare PCIe power cables →rule out a flaky cable as the cause of crashes or black screens
The step-by-step test
1. Inspect the card cold, before powering anything
Look at the PCB under good light. Bulging or leaking capacitors, brown/scorched marks near the VRMs, bent PCIe fingers, or a sagging/cracked die area are instant walk-aways. Spin each fan by hand — they should turn freely and quietly, with no grinding. Check that the backplate screws are present and the heatsink isn't visibly loose. A card that was mined on for years often has caked dust and dried-out, crusty thermal paste at the fan hubs.
2. Read the real temps with HWiNFO64 first
Install HWiNFO64 (free) and open Sensors. At idle on the desktop the GPU should sit roughly 30-45°C. Note the idle hotspot/junction temp too — if 'GPU Hot Spot' is already 15°C+ above the core temp at idle, the thermal paste or pads are degraded. Memory junction temp matters on GDDR6X cards (3080/3090-class): under load it should stay under ~95-100°C.
3. Stress the core: OCCT or FurMark, 10-15 minutes
Run OCCT's GPU 3D test (or FurMark) at the monitor's native resolution for at least 10 minutes — long enough for temps to plateau. Watch for the core temp to stabilize (a healthy air card lands ~65-83°C; over ~85-90°C sustained suggests bad paste/pads or a clogged sink). Crucially, watch the rendered image: ANY flickering dots, colored speckles, scattered triangles, or texture corruption ('artifacts') means dying VRAM or a failing core. Artifacts under load = do not buy.
4. Confirm clocks don't collapse (thermal throttling)
In HWiNFO, watch the GPU core clock during the stress run. A healthy card holds near its boost clock or steps down only slightly. If clocks crater (e.g. from 1800 MHz to 1200 MHz) and temps spike, it's throttling — usually overheating from dried paste. That's fixable with a repaste, but it's leverage to negotiate, not a card to pay full price for.
5. Test the memory specifically
Core stress tests don't always hammer VRAM. Run a memory-focused check: OCCT has a VRAM test, or use a real game at max texture settings for 20 minutes. Mismatched/failing VRAM shows as crashes-to-desktop, driver timeouts (TDR), or artifacts only in memory-heavy scenes. On 30-series cards especially, failing memory is the most common silent fault.
6. Verify it's not a fake or reflashed BIOS
Run GPU-Z. Confirm the GPU name, memory size, and bus width match what's advertised — scammers sell rebadged lower cards (e.g. a 1050 flashed to report as a 1080). Click the GPU-Z BIOS-validity checkmark; a card with a sketchy/unvalidated VBIOS is a flag. The 'Default Clock' vs 'Boost' should match reference specs for that model.
Red flags — walk away if you see these
- Any visual artifacts (speckles, colored dots, corrupted textures) during the stress test
- Core temps over ~90°C or clocks collapsing under sustained load
- Bulging capacitors, scorch marks, or a sagging/cracked die
- GPU-Z reports a different chip/memory than advertised
- Seller refuses to let you run a 10-minute stress test in person
See graphics card GPU listings on eBay → (affiliate)
FAQ
- How long should you stress test a used GPU?
- At least 10-15 minutes — long enough for temperatures to plateau. Most thermal and memory faults appear within the first 10 minutes of sustained 100% load.
- Is FurMark or OCCT better for testing a used GPU?
- Both work. OCCT is gentler and has a dedicated VRAM test; FurMark pushes the hottest sustained load (a 'power virus'). Run OCCT for general health plus a VRAM pass; use FurMark briefly to find the worst-case temperature.
- What temperature is too hot for a used GPU?
- Under sustained load, an air-cooled card should stabilize roughly 65-83°C. Over ~85-90°C, or memory-junction over ~100°C, points to degraded thermal paste/pads — fixable but use it to negotiate.
- Is it safe to buy a GPU that was used for mining?
- Not automatically a dealbreaker — mining cards often ran cool at steady clocks. But check fan bearings (they spin 24/7) and expect dried thermal paste. Stress-test for artifacts and budget for a repaste/refan.
These are practical buyer checks, not a professional appraisal. For high-value items, get an expert opinion before paying.