Browser GPU Performance Test
A GPU is a device that draws graphics, but it is also a device that handles many calculations at the same time. 3D scenes, maps, video effects, photo editing and some AI features in the browser all get help from the GPU. Web Bench looks at this in two branches, graphics and compute. On the graphics side, it renders large numbers of triangles, 131,072 particles and bloom and blur post-processing off-screen and measures the average FPS. On the compute side, it runs matrix multiplication, image convolution, particle simulation and parallel sums, and also verifies that the result buffers are correct. Because nothing is shown on screen, the test is not capped by your monitor's refresh rate, which makes it easier to compare how much work the GPU actually did. Supported features keep changing with the combination of browser, operating system and driver, so the starting point is to check what actually runs in your current environment. Even on the same device, switching browsers can change the GPU path, so noting which browser and which path you used makes the results much easier to interpret later.
GPU tests use WebGPU first, and if it is unavailable, graphics tests fall back to the WebGL2 path. Compute tests require WebGPU, and the f16 matrix multiplication runs only when the shader-f16 feature is supported. This means that even GPUs with the same name can run different tests depending on the browser and driver. Unsupported tests are treated as skipped rather than as errors, and the reason is shown on the results screen. When comparing, first check whether the graphics path of the two results is the same. Placing WebGPU and WebGL2 results on the same line and ranking them can be misleading. If WebGPU is turned off, see the guide on how to turn on WebGPU. Checking the skip reason on the results screen tells you which feature was missing, and that can also be a clue for fixing the problem.
GPU measurements are strongly affected by heat. Many devices score high in a short test and then drop when the load continues for a long time. If you want to see this difference, run the GPU load in the stress test for 5 or 10 minutes and watch how the score changes. On a laptop, the GPU power limit also changes depending on whether it is plugged in and which power mode is selected. Updating your graphics driver and browser before testing, and confirming that hardware acceleration is on, can reduce unnecessary variation. Because the numbers are reference values that vary by environment, they are most useful for comparing before and after you change a setting on the same device. Also remember that the curve from the stress test is only a clue to your device's cooling conditions and is not a direct reading of a temperature sensor. We recommend starting with the quick test to check the representative items, then widening to the standard test when needed.
Start measurement ↗