GPU Stability Test with Error Detection
Most stress tests only tell you how fast a GPU runs. This one also checks that the answers are right. It runs a fixed calculation again and again, reads samples back and compares them with a known-good result, so a card that is fast but producing wrong pixels shows up as errors instead of a good score.
GPU Integrity Check
This canvas shows the animated graphics workload used to stress the GPU. It is decorative; results are listed in the metrics below.
- FPS
- –
- 1% LOW
- –
- FRAME TIME
- –
- SCORE
- –
- TIME
- 0:00
- LOAD
- –
- ERRORS
- 0
Before you start
GPU stress tests can significantly increase GPU load, temperature, power draw and fan speed.
Stop the test if your system becomes unstable, overheats, crashes or behaves unexpectedly. Do not run extreme tests on hardware with known cooling problems. This website cannot guarantee hardware safety.
This is a deliberately punishing test. It can push your GPU to its thermal and power limits for minutes.
- Laptops: plug in the charger and make sure the vents are clear.
- Make sure your GPU is cooled properly. Do not run this with known fan or cooling problems.
- Save your work and close other important tabs. A weak or overheated GPU can make the screen flicker or the graphics driver reset, which may close apps.
- You can stop at any time with the Stop button or the Escape key. The test also stops itself if frames take more than two seconds.
Long run: this test keeps the GPU busy for five minutes or more, so your computer will get warm. Laptops should be plugged in. You can stop at any time with the Stop button.
About this tool
What the test checks
A deterministic shader produces output we can predict exactly. Sample tiles are read back and compared with the first pass and with a reference calculated on your CPU. A separate memory check writes bit patterns into video memory, waits, and verifies them. Every mismatch is counted with a timestamp.
What it cannot see
A web page cannot read temperature, clock speed or power, and WebGL cannot exercise tensor or ray-tracing hardware. The test only detects wrong results it can read back, and it does not say why they happened. If you get errors, repeat the run at stock settings and with other software before drawing conclusions.
How to read the results
- 0 errors: PASS for this run. It does not prove the GPU is stable forever.
- 1 or more errors: FAIL. The result is rated tier D and shown as Unstable, with advice to stop overclocking or heavy use until you have investigated.
The table also lists the number of passes (stacked load) and the highest load step reached.
Running it safely
Start at the default intensity. Torture stacks eight passes of the work, so it is much heavier. Keep the Stop button in view, close other heavy programs and avoid it on a laptop without airflow. The test stops itself if frames take longer than two seconds.
Frequently asked questions
Can this prove my GPU is faulty?
No. It reports errors it measured. A failed run is a strong reason to look further, but it can come from an overclock, a driver, or a hardware fault, and the test cannot tell which.
Is zero errors a guarantee?
No. It means no mismatches were seen in this run, in the data the browser let us read back.
Will it damage my GPU?
It is a normal graphics workload and the load is bounded, but any heavy load makes a PC hot. Stop it if the machine is hot or loud.
What if I get one error?
Treat it as a warning. Run it again at a lower intensity and restore any overclock to defaults. Repeated errors point to instability.
Share feedback about this tool
Did it work for you? Tell us your GPU or device, your browser and what you saw. Your email address is never shown.