What the CPU test measures
The CPU score answers a narrow but useful question: how quickly this browser completes the same short arithmetic loop with one worker and with a small worker group. The counter represents completed inner-loop iterations, not individual processor instructions or a clock-frequency reading.
Use the single-worker figure to notice changes in one stream of browser work and the parallel figure to see how this workload scales across the workers made available to the page. The score combines both rates only for a convenient same-test baseline.
What the GPU test measures
The GPU score answers whether a simple accelerated WebGL path is available and how much synchronized draw throughput that path delivers during a short run. Because every draw uses the same compact shader and off-screen surface, the number is a repeatable browser signal rather than a simulation of a game.
Read the capability report beside the score. WebGL version describes the active API, WebGPU availability is a separate compatibility signal and a generic renderer name may simply reflect browser privacy protection. None of those fields alone ranks the physical graphics card.
Create a fair comparison
Use the same browser version, power mode and device temperature. Connect a laptop to power when comparing maximum performance, close demanding apps and keep the tab visible. Let an unusually hot system cool, then run at least three times and use the middle result.
Change one condition at a time. If testing a browser update, leave hardware and power settings unchanged. If testing a power profile, keep the browser and room conditions fixed. Record the date and reported worker or graphics context information with the score.
- Stabilize power, temperature and background activity.
- Run three times in the same current browser and keep the median.
- Change one variable, repeat the same process and compare percentages rather than tiny point differences.
Why scores change between browsers and runs
Browsers use different JavaScript compilers, security policies, graphics translation layers and scheduling decisions. A driver update can alter WebGL behavior; a browser may select an integrated GPU on battery; privacy protection can hide the renderer without changing acceleration.
Operating systems continuously schedule background work, adjust clock rates and manage heat. Short tests are intentionally convenient but therefore sensitive to brief interference. A small difference is ordinary noise; a large repeatable difference under controlled conditions deserves investigation.
Use the right tool for the decision
Use these browser tests to verify that browser computation and accelerated graphics work, compare a setting on the same machine, or identify a large regression. Use application-specific native benchmarks when deciding how a computer will perform in a game, renderer, compiler or scientific workload.
Scores from different benchmark products are not interchangeable because the operations, durations and scaling formulas differ. The complete formulas and workload details for this site are maintained on the methodology page so a score can be interpreted without guesswork.