cuda-oxide compiles SIMT kernels directly to PTX, while cutile-rs brings Tile-based programming to stable Rust.

NVIDIA introduced CUDA Rust, a toolchain that compiles GPU kernels written in Rust to PTX. The project offers two tracks: cuda-oxide for SIMT-style kernels and cutile-rs for Tile-based GPU programming. Both paths enforce memory safety at compile time.

NVIDIA’s push into Rust spans the Nova Linux driver, Dynamo’s Rust core, and NVTX Rust bindings. CUDA C++ and CUDA Python remain mature toolchains, but the company said it will mature CUDA Rust into 2027 and beyond. CUDA Rust closes the gap by compiling kernels written in Rust to PTX.

Hugging Face’s Grout inference engine and mistral.rs already use the cutile-rs crate from crates.io. cutile-rs runs on stable Rust 1.89 or later with CUDA 13.3 and no custom LLVM. cuda-oxide remains in early alpha and requires a pinned nightly toolchain and LLVM.

NVIDIA plans to grow and mature CUDA Rust into 2027 and beyond. The roadmap includes inter-language interoperability among CUDA Rust, CUDA C++, and CUDA Python so a frontend choice does not lock developers out of other ecosystems. Developers can run the cuda-oxide SIMT example and the cutile-rs hello_world example today.

What matters

• NVIDIA introduced cuda-oxide and cutile-rs as two tracks for writing GPU kernels natively in Rust.
• Both projects enforce compile-time memory safety and let developers write GPU kernels without leaving Rust.
• NVIDIA plans interoperability among CUDA Rust, CUDA C++, and CUDA Python in future releases.

Why it matters

NVIDIA plans interoperability among CUDA Rust, CUDA C++, and CUDA Python in future releases.

This GenAI News article was prepared in original wording using reporting and materials published by NVIDIA Developer Blog. Source reference: https://developer.nvidia.com/blog/introducing-cuda-rust-two-tracks-for-writing-gpu-kernels/.

Drafted by the GenAI News review pipeline.