Microbenchmarks

A collection of micro-benchmarks comparing Julia's performance against other languages.

Benchmarks

All benchmarks implement identical algorithms across languages (serial, single-core). Times are normalized relative to C.

Benchmark	Description
`iteration_pi_sum`	Alternating power-series summation (nested loops)
`recursion_fibonacci`	Doubly-recursive Fibonacci(20)
`recursion_quicksort`	Quicksort on 5,000 random numbers
`parse_integers`	Parse 1,000 random hex strings to integers
`print_to_file`	Write 100,000 formatted lines to /dev/null
`matrix_statistics`	Statistics on random 5x5 matrices (1,000 iterations)
`matrix_multiply`	Multiply two random 1,000x1,000 matrices (BLAS)
`userfunc_mandelbrot`	Mandelbrot set computation over a grid

Mathematica and Matlab benchmarks are available but not run in CI.

Two workflows run the benchmarks:

benchmarks-single-runner.yml produces the published results. It runs every language on one runner: it installs and compiles everything first, then runs the whole suite three times back-to-back and keeps the best time per benchmark. Its report job builds and deploys the results site from those results, which are also attached to the run as the bench-all artifact and shown in the job summary.
benchmarks.yml runs each language in its own job, as a quick per-language check. Jobs can land on runners with different CPUs, so its times are not comparable across languages and are not published.