Nvidia Announces CUDA Rust for Native GPU Programming
Nvidia announced CUDA Rust, two tracks for writing GPU kernels in Rust: cuda-oxide for SIMT and cutile-rs for Tile-based programming.
According to Nvidia's technical blog, the company announced CUDA Rust, two tracks for writing GPU kernels natively in Rust. The two projects are cuda-oxide (SIMT track) and cutile-rs (Tile track).
The mechanism: cuda-oxide is a custom rustc codegen backend that compiles SIMT-style GPU kernels written in Rust directly to PTX using the Pliron IR framework and LLVM. It requires a pinned nightly toolchain and LLVM. cutile-rs enables Tile-based GPU programming in stable Rust (1.89+), where the compiler manages thread mapping and memory layout through CUDA Tile IR JIT compilation. It runs on stable Rust with CUDA 13.3 and no custom LLVM.
Both projects enforce memory safety at compile time: cuda-oxide uses DisjointSlice and launch contracts to prevent aliasing, while cutile-rs uses tensor partitioning and ownership to guarantee exclusive access. Both require compute capability 8.0 or later.
What is unconfirmed is production readiness: cuda-oxide is early alpha, cutile-rs is further along and published on crates.io, already used in HuggingFace's Grout inference engine and mistral.rs. Coverage is incomplete and APIs will move. Nvidia plans inter-language interoperability between CUDA Rust, CUDA C++, and CUDA Python.
What to watch is whether CUDA Rust matures into a production toolchain by 2027, and whether the Tile track becomes the default for new GPU programming. The Rust safety guarantees at compile time are the key differentiator versus CUDA C++.
Desk take
CUDA Rust gives GPU kernel programmers Rust's compile-time memory safety. The Tile track (cutile-rs) is further along and already used in inference engines. Watch for production maturity by 2027 and whether Tile becomes the default programming model.
Rust's compile-time safety catches aliasing bugs that CUDA C++ misses, which matters for the reliability of GPU kernels in production AI workloads.
Source dispatch
Article URL: https://developer.nvidia.com/blog/introducing-cuda-rust-two-tracks-for-writing-gpu-kernels/ Comments URL: https://news.ycombinator.com/item?id=49724881 Points: 968 # Comments: 404
