Documentation / CPU Benchmark

CPU Benchmark

How it works #

We use a CPU benchmark inspired by Wikipedias CPU benchmark included in the Autonomous Systems performance report. It’s super simple: it’s a loop that we run in the browser after the page has finished loading and our other tests has finished. It produces a metric in milliseconds of how long time it takes to run.

How can you use it? #

There’s a couple of different use cases:

  • Compare the benchmark on a real phone versus emulated mobile phone tests. Run the test on a phone and run it on your computer and compare the value to get a feeling on how much faster (or slower) the tests are on the computer.
  • Keep track of the benchmark over time on the server that runs your tests. Is the benchmark unstable, your other metrics will also be unstable.
  • If you collect performance metrics from real users you can also collect the CPU metric from your users (make sure to do it off the main thread as Wikipedia). That way you can compare your synthetic tests with your RUM data.

Here’s an example of what the variation looks like running on a Moto G4 running the tests in Chrome. We run eleven runs and in the graph you see the min, median and max value. CPU benchmark variation running on a Moto G4

Test page #

We also have a special test page you can use to see the benchmark on your own browser/computer/device without running Access the page and look at the benchmark metric.

You can also use the page to calibrate your CPU throttling rate when you use Chrome. Access the page and look at the result and fine tune your throttling rate. --chrome.CPUThrottlingRate 5 -b chrome