tandem-webgpu

Tandem8x32 for JavaScript: a WGSL shader fills the stream on the GPU, and the same package runs every draw on the CPU for Node, Deno, Bun and browsers. It produces the stream of the specification bit for bit.

• API: the CPU class, the GPU fills, the shaders and parallel use.
• Tests: fixtures, hashes and what each suite checks.
• Performance: CPU and GPU figures and rejected variants.

Install

npm install github:tandem-rng/tandem-webgpu

The install builds dist/ through the prepare script. In a clone, Deno, Bun and Node 24 or later import src/mod.ts directly. The GPU path needs WebGPU. No dependencies.

The stream is bit exact with the specification, and the CPU and GPU return the same values, except GPU normals, which agree to 16 ulps plus 1e-6, and GPU exponentials, which agree to 8 ulps plus 1e-6. CPU exponentials are bit exact with tandem-c, and CPU Float32 normals agree with it to 4 ulps.

AI assistance

This port was written with the help of large language models under human direction. The design and the specification are human work, as is much of the Julia implementation. The code is tested bit for bit against every vector of the specification and against long stream dumps from the Julia implementation, and every value must match. The output does not depend on who or what wrote the code.