pyproc
Web Python speed, measured in the tab
This page uses the public PyProc API only: boot a worker pool,
run a single-worker numpy matrix multiply, then run the same matrix multiply sharded across
4 Python workers. Three warmed samples are summarized as median speedup and p95 latency
before the speed number is accepted.
Booting 4 Python workers with numpy...
4 interpreters
4 independent GILs
Float64 numpy matmul
The API under test:
const os = new PyProc({ packages: ["numpy"], setup: "import numpy" });
await os.boot(4);
await os.matmul(a, b, { parts: 1 });
await os.matmul(a, b, { parts: 4 });