RunTheTests
/

WebGL GPU Benchmark

A graphics score that admits when the display, not the GPU, was the limit.

How this benchmark works

A fragment-heavy scene is rendered off-screen at 512 by 512 for four seconds and the completed frames are counted. Several details keep the number defensible. The shader does real arithmetic rather than filling pixels with a colour, because a trivial shader measures how fast the browser issues draw calls — a property of the browser, not the graphics hardware. Each frame ends with a call that blocks until the GPU has actually finished, so the loop counts frames drawn rather than frames queued; without it, a deep driver queue produces a spectacular figure that means nothing. Rendering happens off-screen so compositing and display presentation stay out of the measurement. And when the result settles near sixty frames per second, the page reports that the display refresh was likely the ceiling rather than claiming a score, because a browser is not permitted to present frames faster than the compositor allows.

What the results mean

Relative index
Rendering rate against a reference machine, where 100 is that machine. A readable scale for this page and nothing more — not an industry standard.
Frames per second
Completed frames of the test scene per second, each one waited on until the GPU finished it. Not the frame rate of any real application, which depends entirely on what that application draws.
Limited by the display refresh rate
Shown when the rate lands close to a standard refresh. It means the honest conclusion is that the GPU is at least this fast; how much faster cannot be established from a browser.
Renderer / vendor
The graphics device as the driver names it. Most browsers now mask these to limit fingerprinting, and where they are hidden the page says so rather than substituting a guess from the user agent.

Common problems and fixes

The score is very low on a machine with a good graphics card
Check that hardware acceleration is enabled in the browser. With it off, everything is rendered on the processor in software, which produces a result an order of magnitude below the same machine's capability.
A laptop scores far worse than expected
It is probably rendering on the integrated graphics. Laptops with two graphics processors route browsers to the low-power one by default, and the graphics control panel is where that is changed — usually per-application.
WebGL is unavailable
That happens with hardware acceleration disabled, inside virtual machines, and in browsers with WebGL switched off for privacy. It is a configuration state rather than a fault in the graphics hardware.
The renderer shows as masked
That is expected and deliberate. Browsers restricted the extension that exposes it because it is a strong fingerprinting signal. Your system's own display settings will name the device.

Frequently asked questions

Why does the result stop at around 60?

Because a browser presents frames in step with the display, and on a 60 Hz screen that is the ceiling regardless of how much headroom the GPU has. The workload is sized to stay below it on typical hardware, and when the result lands there anyway the page reports the ceiling rather than pretending the number is a measurement of the graphics card.

Can I compare this against a 3D benchmark score?

No. Native benchmarks drive the graphics hardware directly with compiled code and their own frame pacing. This runs through a browser, a compositor and a driver path built for web content. The numbers are not on the same scale and no conversion exists.

Why is the renderer name hidden?

Browsers restricted the WebGL debug renderer extension because the exact driver string is a strong fingerprinting signal — it narrows a visitor down considerably on its own. Where the browser hides it, this page reports it as masked rather than guessing from the user agent, which would be wrong often enough to mislead.

What is this benchmark good for?

Comparing a machine against itself. Before and after a driver update, with and without hardware acceleration, on integrated graphics versus discrete — those comparisons are real and attributable. Comparing your number against someone else's is not, and the page does not offer a way to.

More in Device