Benchmark builder
Easily compare execution time of tasks with this very simple online benchmark builder.
About Benchmark builder
Compares sets of measurements you have taken yourself, for example execution times of two implementations: enter the values for each suite and get the mean and variance, ranked, with a Markdown table ready to paste into an issue or a pull request.
When you would use it
- Presenting timings of two or three code variants in a pull request description.
- Summarising repeated manual measurements (page load, build time, request latency) without a spreadsheet.
- Checking whether a difference between two approaches is consistent or just noise.
How to use it
- Name each suite and type its measurements, one number per field; add suites with the plus button.
- Set the unit (ms, s, MB) so the table is labelled.
- Read the mean and variance, then copy the result as a Markdown table or as plain text.
Things worth knowing
- The variance shown is the population variance (mean of squared deviations). A large variance relative to the mean means the measurements are noisy, and a small mean difference between suites may not be meaningful.
- Take at least five measurements per suite, discard the first run (cold caches, JIT warm-up) and measure the variants under the same conditions, or the comparison is not fair.
- The tool does not run any code. It only aggregates the numbers you provide; use your language's benchmarking library to produce them.
- Suites and the unit are remembered in this browser, so you can come back and add measurements later.
Frequently asked questions
- Why does my faster variant sometimes measure slower?
- Background activity, CPU frequency scaling and garbage collection all add noise. More repetitions and reporting the mean with its variance (or better, the median) make the comparison robust.
- Can I paste a list of numbers at once?
- Not currently; values are entered one per field. For long series, aggregate them in your benchmarking tool and enter the summary values here.