NVIDIA Launches CUDA Rust with Two Paths for Writing GPU Kernels

NVIDIA announced CUDA Rust on its official blog: developers can now write CUDA GPU kernels natively and safely in pure Rust. The project was promoted with the tagline "CUDA Rust is here" and is open-sourced on GitHub. This marks a major official endorsement of the Rust community within the CUDA ecosystem.

Confirmed

• Two paths are offered: cuda-oxide and cutile-rs.
• cuda-oxide targets the SIMT programming model via a custom rustc codegen backend, compiling SIMT-style Rust kernels directly to PTX using the Pliron IR framework and LLVM; it requires the nightly toolchain.
• cutile-rs targets the Tile programming model.
• The project was released by npmashi et al., with code available on GitHub.

Why it matters

• Combining Rust's memory safety with GPU programming can reduce security risks in kernel development and opens the CUDA ecosystem to developers unfamiliar with C++.
• The dual-track design (SIMT and Tile) spans from fine-grained thread-level to highly abstracted tile-level programming, reflecting NVIDIA's support for diverse GPU programming paradigms.

2026-09-09 ~ 2026-09-10 · 5 related posts

• [source] NVIDIA unveils CUDA Rust: cuda-oxide and cutile-rs, two tracks for writing GPU kernels — ssh4net · 2026-09-09
• NVIDIA launches CUDA Rust: two tracks to write GPU kernels in Rust — ducha_aiki · 2026-09-09
• You can now write CUDA kernels in plain Rust via cuda-oxide and cutile-rs — rjurney · 2026-09-10

2 near-duplicate retellings: blelbach · blaizedsouza