How to test a GPU before selling it
About half an hour with free tools tells you whether a graphics card is healthy — and it is the same set of checks any serious buyer runs when the card arrives. Do them first and there are no surprises about the offer.
Why test before you sell
A card that looks fine on the desktop can still fail under load: artifacts, driver crashes or a black screen appear only when the GPU and its memory are working hard. Finding that out before you ship saves you the trip, and a card that passes can be described honestly as tested and working.
1. Confirm the card and its memory
Install GPU-Z, a free utility. Its main tab names the GPU, the memory size and type, and the BIOS version. Check the memory size against the box: several cards were sold with two memory sizes under almost the same name, and the offer differs. While you are there, make sure the latest driver from NVIDIA, AMD or Intel is installed, so a failure in the next step is the card and not an old driver.
2. Run a stress test
A stress test holds the GPU at full load for long enough to expose instability. Run one for 15 to 30 minutes:
- 3DMark’s stress tests (Time Spy or Steel Nomad) loop a benchmark twenty times and report frame-rate stability; a healthy card passes at 97% or higher.
- FurMark or Unigine Superposition are free alternatives that load the GPU heavily.
- A demanding game at high settings for half an hour is a fair real-world check too.
Watch for sparkles, coloured blocks or lines on screen, the driver resetting, the screen going black, or the computer restarting. Any of those is a fault.
3. Test the video memory
Memory faults often show only as occasional artifacts, so test the memory on its own. OCCT has a dedicated VRAM test, and memtest_vulkan is a free, open-source one for Windows and Linux. Let either run for ten minutes or so; any reported error means a faulty memory chip.
4. Watch the temperatures
Keep HWiNFO or GPU-Z’s Sensors tab open during the stress test. Three readings matter:
- GPU temperature. Most cards run in the 70s °C under a long load, some a little higher. A card that climbs until it throttles and loses clock speed needs attention.
- Hot spot. The hottest point on the chip. A gap between it and the GPU temperature much above 20 °C usually means the thermal paste has dried out or the cooler isn’t seated evenly.
- Memory junction, on cards that report it. Readings pushing past 100 °C usually point to worn thermal pads, common on cards that were mined on.
High temperatures don’t make a card faulty on their own, but they tell you what a buyer will see.
5. Check fans, outputs and the power connector
- Fans: under load every fan should spin, without grinding or rattling. Many cards stop their fans at idle on purpose, so check during the stress test, not on the desktop.
- Outputs: plug a monitor into each DisplayPort and HDMI port in turn. A dead output is easy to miss if you only ever used one.
- Power connector: on cards with the 16-pin connector (used on many RTX 40 and 50 cards), look for melted, browned or deformed pins on both the card and the cable. Any sign of heat damage counts as a fault.
- The board: look for cracks, lifted components or burn marks, and for a sagging card whose board has bent near the slot.
If it fails
We only pay for cards that work, so a card that fails these checks gets no offer from us. If you’re shipping other parts, put it in the box and we’ll recycle it responsibly. If yours passes, you know exactly what you’re selling — pack it well and see what it’s worth. We run the same kind of checks at inspection; if we find something you didn’t, you see the reason in a revised offer and can have the card back free.