How to interpret your results

Start with the conditions

A useful result is a number plus the conditions that produced it. Record the renderer name, browser version, duration, texture dimensions, texture count and filtering mode. Check whether the renderer is the GPU you expected. A laptop may select integrated graphics, and a virtual machine may expose a virtual or software renderer.

Before comparing two runs, ask whether the tab stayed visible, the power mode stayed the same and other programs were closed. If any of those changed, repeat the run under matching conditions. Comparing a phone with a desktop using this workload does not produce a reliable overall GPU ranking.

Read the metrics in context

These measurements describe this page’s rendering behavior. None is a count of corrupted memory bytes.

A worked example

Suppose three visible-tab runs with four 1024-pixel textures show 59, 60 and 59 average FPS. The texture data estimate remains 16 MiB in all runs. This suggests the workload is repeatable at approximately the display scheduling rate; it says nothing about untested memory.

Now switch to four 2048-pixel textures. The data estimate becomes 64 MiB. If repeated runs fall to 42–45 FPS, record that setting-dependent difference. It could involve sampling cost, caches, browser scheduling or other system work. It does not by itself demonstrate faulty VRAM. If returning to the original settings also stays slow, check background work, power settings and temperature using suitable system tools.

These numbers are illustrative examples, not measured results from a particular graphics card. The site does not publish a cross-device leaderboard or claim a validated score distribution.

Why 60 FPS and 100% are not certificates

requestAnimationFrame typically follows the display refresh cycle. On a 60 Hz display, the browser can deliver approximately 60 callbacks per second even if the GPU could render this scene much faster. Higher refresh rates, power saving and browser behavior change that ceiling.

The consistency index measures variation, so a slow but regular sequence can still look consistent. The performance score uses a fixed 60 FPS target. Do not use either percentage or star rating to certify an overclock, accept a warranty claim or approve a used card purchase.

Save a useful comparison

Export Results creates a JSON file on your device. It contains the displayed settings, metrics, time and browser/system information. Review the file before sending it to anyone: renderer, user-agent and screen details can reveal information about your device.

Use a small log with one row per run: date, settings, average FPS, frame consistency, foreground/background and notes. Compare repeated runs on the same setup. If the actual symptom is corruption or crashing, use the troubleshooting and used-GPU guides for the next diagnostic step.