Benchmarks 12 numerical/linear-algebra operations across 15 language implementations, measuring speed and memory on Apple Silicon (macOS).
| Language | Library |
|---|---|
| Python (NumPy) | NumPy + SciPy + system BLAS |
| Python (Pandas) | Pandas (delegates to NumPy for decompositions) |
| Python (Polars) | Polars (delegates to NumPy for decompositions) |
| Python (JAX) | JAX / XLA (JIT-compiled) |
| Python (Numba) | Numba JIT |
| Octave | Built-in BLAS/LAPACK |
| R | R + BLAS |
| Julia | LinearAlgebra.jl + FFTW |
| Rust | nalgebra + realfft |
| C++ | Eigen |
| C | CBLAS + LAPACK + vDSP (Accelerate) |
| Fortran | BLAS + LAPACK (Accelerate) |
| Haskell | hmatrix (BLAS/LAPACK) |
| Swift | Accelerate |
| Go | gonum |
Matrix Multiply · Matrix Inverse · LU Decomposition · Eigenvalue Decomp · Cholesky · SVD · Linear System Solve · Vector Dot Product · Hadamard Product · QR Decomposition · FFT (real, 1M) · Sort 10M floats
Prerequisites: macOS on Apple Silicon, pip install -r requirements.txt, plus whichever of the other toolchains you want (Octave, R, Julia, Rust, Eigen, GHC/cabal, Go, gfortran; Swift and clang come with the Xcode Command Line Tools). Languages whose toolchain is missing are skipped. run_all.sh auto-installs the Julia packages JSON3 and FFTW and the R package jsonlite if they are missing.
./run_all.sh # run all languages and regenerate report
python3 generate_report.py # regenerate report from existing resultsresults/*.json and index.html (the GitHub Pages report) are tracked in git; re-run and commit both together.
See CLAUDE.md for per-language build instructions.