RUST · GPU · HIGH-PERFORMANCE COMPUTING

From kernel.
To model.

One Rust stack for GPU kernels, compute libraries, tensors, and models. Build at the level you need. Stay connected to the layers below.

cargo add ruda --features cuda

Rust 2024Direct PTXNative PyTorch

RUDA / COMPUTE STACK+

Models & trainingruLLM / ruda-nn / ruda-optim

Tensors & frameworksruda-tensor / ruda-autodiff / ruda-torch

Compute librariesruBLAS / ruDNN / ruFFT / ruTENSOR

Kernels & executionruda-kernel / ruda-compiler / runtime

GPU KERNEL → TENSOR → MODEL

01 / BUILD

Your starting point.

Write a kernel, train a model, or bring native operations to PyTorch.

Explore the Rust kernel DSL, execution model, and direct PTX path.

Run on ruda:0 with native operators, streams, and first-order training.

Prepare model weights and explore text and image inference with ruLLM.

02 / COMPUTE

Low-level control. High-level composition. Clear boundaries between every layer.

03 / GET STARTED

Clone the workspace, then follow the environment guide for your GPU and toolchain.

● ● ●TERMINAL

git clone https://github.com/shuqi2077/RUDA.git
cd RUDA

The NVIDIA path needs Rust/Cargo, a linker, an NVIDIA driver, and the CUDA Toolkit.