NVIDIA launches CUDA Rust: two paths to write GPU kernels natively in Rust
blelbach · x · 2026-09-17
NVIDIA introduced CUDA Rust, enabling developers to write GPU kernels natively in Rust rather than just launching them from it, via two tracks:
- cuda-oxide: a custom rustc codegen backend compiling SIMT-style kernels directly to PTX through the Pliron IR framework and LLVM; early alpha, requires a pinned nightly toolchain and custom LLVM.
- cutile-rs: Tile-based programming on stable Rust (1.89+ with CUDA 13.3), where the compiler manages thread mapping and memory layout via CUDA Tile IR JIT compilation; published on crates.io and already used in HuggingFace's Grout inference engine and mistral.rs.
Both enforce compile-time memory safety — cuda-oxide uses DisjointSlice and launch contracts to prevent aliasing, while cutile-rs relies on tensor partitioning and ownership. NVIDIA also plans inter-language interoperability between CUDA Rust, CUDA C++, and CUDA Python.
More from Infra
- How Bell Labs Missed the Microchip: IEEE Spectrum Revisits a Landmark Tech-History Blunder — ArtificialOther · 2026-09-17
- Agentic AI systems are the next network users: 40% of enterprise apps to include agents by 2026 — seankinneyRCR · 2026-09-17
- B200 spot rental up 80% in 8 months as demand outpaces compute buildout — JOBhakdi · 2026-09-17
- MLX-Serve 26.9.3 ships: Qwen Flash Next tops 100 tok/s on M4/M5 Max Macs — TheMoonMidas · 2026-09-17
- Running a 14B Model on 16GB RAM: 'My PC Is a Toaster Now' — Aggravating_Site381 · 2026-09-17
- fal engineering head: we'll never pre-train, inference compute is the real moat — jfischoff · 2026-09-17