How to Compare Two GPUs with the Same Browser Workload
Use a controlled comparison log for WebGL tests, avoid unverified vendor scores, and account for browser, driver, and display limits.
Comparing two graphics cards requires comparable observations. A browser workload can answer a narrow question about a scene and configuration, but it cannot establish an overall NVIDIA-versus-AMD ranking. This site has no measured vendor leaderboard or laboratory dataset.
Start with the question
For browser graphics, ask how the same scene performs on two recorded systems. For a specific game or compute application, test that application instead. A card may behave differently across those workloads, so avoid turning a single FPS value into a purchasing verdict.
Hold the comparison conditions fixed
Use the same browser version, OS where practical, driver configuration, viewport, zoom, shader code, camera, and power mode. Record any difference that cannot be controlled. Keep both devices ventilated and run a brief warm-up before the observation window.
The volume shader demo is editable. Retain exactly the same shader source for both systems. The instancing test uses a simple triangle workload; it is a different experiment, not a second score on the same scale.
Use a blank log instead of invented scores
| Condition or observation | System A | System B |
|---|---|---|
| Adapter and renderer, if exposed | Record | Record |
| Browser, OS, and driver version | Record | Record |
| Viewport, zoom, shader revision | Record | Record |
| Workload settings and duration | Record | Record |
| FPS observations from three repeats | Record | Record |
| External temperature monitor, if used | Record | Record |
This table is a template. It contains no real benchmark result or evidence that one vendor leads.
Check whether the display is limiting the result
If both runs stay near their display refresh limits, similar FPS does not demonstrate equal compute capacity. If you change resolution to make a shader more demanding, repeat the same change on both systems and document the new workload.
A renderer string can be unavailable due to privacy settings. It may also describe a browser translation layer. Record it as browser evidence, not as a complete hardware inventory.
State a bounded conclusion
Report the workload and conditions alongside the difference. Explain uncontrolled variables and repeatability. Do not infer dedicated ray-tracing performance from our WebGL ray demo, or memory bandwidth from a texture allocation counter.
See renderer extension documentation and our measurement methodology.