GPU Benchmark (WebGL)

How to run the GPU benchmark

1. Close games, video editors and other heavy tabs so they don’t compete for the graphics card. On a laptop, plug in the charger: many laptops slow their GPU down on battery.
2. Check the Renderer line. It shows the graphics card the browser is using, such as an integrated Intel or AMD GPU or a dedicated NVIDIA or AMD card.
3. Press Start benchmark and keep the tab in front. The test runs seven stages of about three seconds each, around 25 seconds in total.
4. Watch the stage name and the progress bar. Press Stop at any time; the score then covers only the finished stages.
5. Read the Score and the Results per stage table. Run the test two or three times and keep the best result: the first run is often lower while the GPU raises its clocks.

What the benchmark measures

The first five stages render a procedural landscape with a raymarching fragment shader. For every pixel of a 1280 × 720 image, the shader marches a ray step by step across terrain made of five octaves of value noise. The load doubles from stage to stage: 16, 32, 64, 128 and 256 steps per pixel, so the last stage computes about 236 million steps per frame. Every pixel always runs the same number of steps, so the work grows exactly with the load. These stages test raw shader throughput, the main factor in modern game performance.

The Draw calls stage draws 2,000 small squares per frame, each with its own draw call and colour. It measures how quickly the browser, the driver and the CPU can issue commands, which limits scenes with many separate objects. The Triangles stage draws one million tiny triangles per frame from a static vertex buffer, to test geometry throughput.

How the numbers are computed

After drawing each frame, the tool reads back one pixel. This forces the browser to wait until the GPU has really finished, so the Median frame time is the real rendering time and is not capped by your screen’s refresh rate. The first frame of each stage is discarded because it includes shader warm-up.

• FPS (uncapped) = 1000 ÷ median frame time in ms.
• Throughput = work per frame × FPS: pixel-steps, draw calls or triangles per second.
• Score = 1000 × geometric mean of each stage’s throughput divided by a reference value (10 billion pixel-steps/s, 500,000 draw calls/s and 500 million triangles/s).

A score of 1000 is roughly a recent integrated GPU; dedicated graphics cards usually score several times higher. The score is only meant for comparing devices, browsers and settings with this tool. It is not comparable with native benchmarks, because WebGL adds a translation layer (ANGLE on Windows and macOS) and browsers limit some features.

If one frame takes more than a second, the heavier shader stages are marked Skipped (too slow) to avoid freezing the computer or triggering a driver reset.

When results look wrong

If the renderer contains “SwiftShader”, “llvmpipe” or “Basic Render”, hardware acceleration is off and WebGL runs on the processor. The tool then lowers the resolution to 640 × 360, and the results reflect your CPU. Turn on hardware acceleration in the browser settings and update the graphics driver. On laptops with two GPUs, make sure the browser uses the dedicated one in the system’s graphics settings.