GoKawiilOpenBLAS now runs in WebAssembly, speeding up NumPy in browsers up to 31x
Emscripten-forge has added OpenBLAS support to its NumPy package for WebAssembly, replacing the previous unoptimized fallback loops used for matrix operations in browser-based Python. Benchmarks show np.matmul on 1024x1024 matrices running roughly 30.92x faster for float32 and 14.90x faster for float64, with an upcoming OpenBLAS release and an optional Relaxed SIMD build promising further gains.
GoKawiil's interpretation of the reporting above, not reported fact.
Scientific Python libraries have long been hobbled in the browser because compiled numerical backends like BLAS and Fortran-based LAPACK couldn't easily target WebAssembly, forcing workarounds like f2c translation. Emscripten-forge's language-agnostic approach, building compilers and shared libraries as conda-style packages, could let performance-critical scientific code run natively in browsers across Python, R, and other languages rather than one ecosystem at a time.
- Emscripten-forge's NumPy package now links OpenBLAS for hardware-accelerated matrix math in WebAssembly.
- Matrix multiplication benchmarks show up to 30.92x speedup for float32 and 14.90x for float64 at n=1024.
- Emscripten-forge differs from Pyodide and WebR by being a language-agnostic distribution covering Python, R, C++, Fortran, and more.
Source: notebook.link — Julien Jerphanion, 2026-09-17
Published there as: “Faster NumPy in the Browser”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.