@node-3d/cuda

This is a part of Node3D project.

CUDA bindings for Node.js.

npm install @node-3d/cuda

This package exposes the CUDA driver API pieces used by Node3D examples:

Prebuilt addon binaries: Windows x64/ARM64, Linux x64/ARM64. Standard Windows x64 and Linux builds target CUDA 12.9. Windows ARM64 builds target CUDA 13.4 because NVIDIA introduced Windows ARM64 target support there. CUDA toolkit/runtime and an NVIDIA driver are still required on the host system.

Platform Notes

CUDA does not work on macOS. NVIDIA does not ship a current macOS CUDA runtime, so there is no macOS addon binary and CUDA calls are not expected to succeed there.

The package can still be installed and imported on macOS for cross-platform codebases that share one module graph across Windows, Linux, and macOS. On macOS, basic import and device-count queries are safe, but CUDA operations throw a clear unsupported-platform error. Branch before constructing CUDA objects or allocating memory:

import { getDeviceCount, Device } from '@node-3d/cuda';

if (getDeviceCount() > 0) {
	const device = new Device(0);
	console.log(device.name);
}

• Device, Ctx, Modulex, Function, and Mem native wrappers.
• Memory helpers such as memAlloc, memAllocPitch, and memVBO.
• Module loading and runtime compilation helpers.
• launch and prepareArguments for typed kernel argument setup.
• Selected Thrust helpers used by compute-heavy examples.

import { Ctx, Device, getDeviceCount, memAlloc } from '@node-3d/cuda';

console.log('CUDA devices:', getDeviceCount());

const device = new Device(0);
const context = new Ctx(0, device);
const memory = memAlloc(1024);

memory.free();
context.destroy();

See examples/basic.ts, examples/memory-copy.ts, and examples/vector-add.ts for complete runnable examples. CUDA toolkit/runtime availability is still required on the host system.

Binary Origin

Release archives are built by this repository's public GitHub Actions workflows.

Attestations: https://github.com/node-3d/cuda/attestations

To verify a downloaded archive:

gh release download <tag> -R node-3d/cuda -p <platform>.gz
gh attestation verify <platform>.gz -R node-3d/cuda