A practical checklist for a used GPU

Identify the card before testing it

Check the model label, seller’s description, connectors and the reported model in a trusted system utility. Use the manufacturer’s specifications for the exact model to check advertised memory capacity. Browser texture allocation and renderer text cannot independently verify either the model or its installed memory.

Ask the seller which settings are in use and whether core clocks, memory clocks, voltage, cooling or firmware have been changed. Test with stock settings and save a record of the driver, operating system and model. Do not rely on a claim that a card is healthy because it previously passed this website.

Use the browser run as a quick observation

Begin with a short, modest workload and keep the tab visible. Note whether the renderer is plausible and whether the preview renders without the browser losing its context. Run the same settings again before increasing texture size or count.

This step can reveal a browser compatibility problem or an obvious rendering failure in this environment. It cannot clear a used card for purchase. A clean browser preview covers a small, driver-managed workload without byte verification.

Verify data with a native memory tool

Obtain the utility from the maintainer’s official release source and follow its instructions. memtest_vulkan is one example of a project that describes actual GPU memory verification. Confirm the selected device and write down the requested/tested memory amount, patterns, duration and error results.

Close applications that compete for GPU resources. Start with the utility’s documented defaults. If a utility cannot allocate the requested amount, that alone is not proof of a failed chip: the operating system, display and other processes reserve memory. Do not assume any tool can test every installed byte.

Repeat a reported data mismatch at stock settings and save the log. A reproducible failure is a reason to investigate further or defer the purchase. Do not hide it by lowering clocks and describing the card as fault-free.

Reproduce the intended use

Run an application that represents how you will use the card: the relevant game, render scene or compute task. Observe image correctness, crashes, driver resets, fan behavior and sensor readings with appropriate system tools. A texture test does not reproduce a game’s full shader, compute, power or memory workload.

Keep settings constant and compare a cold run with a repeat run under normal operating conditions. Stop on unusual behavior. There is no universal duration or browser score that guarantees reliability for all applications.

Make the decision from combined evidence

Keep identification details, settings, screenshots, native logs and the seller’s return terms together. A consistent browser run plus a clean native check plus a representative application run provides more useful evidence than any of these alone, but still cannot guarantee future reliability.

If the card fails across multiple workloads at stock settings, defer acceptance and discuss the evidence with the seller or vendor. If only the browser fails, resolve that compatibility issue before inferring a hardware defect. Physical repairs should be assessed by someone qualified to perform them.