Hardware
Introducing CUDA Rust: Two Tracks for Writing GPU Kernels
NVIDIA introduced two Rust projects for GPU development in September 2026: `cuda‑oxide`, a nightly‑only backend that compiles SIMT‑style kernels to PTX via LLVM and uses DisjointSlice and launch contr
NVIDIA introduced two Rust projects for GPU development in September 2026: cuda‑oxide, a nightly‑only backend that compiles SIMT‑style kernels to PTX via LLVM and uses DisjointSlice and launch contracts for compile‑time memory safety; and cutile‑rs, a stable framework that manages thread mapping and memory layout through CUDA Tile IR JIT compilation, also guaranteeing exclusive access with tensor partitioning. Cutile‑rs is production‑ready (already deployed in HuggingFace’s Grout inference engine) and available on crates.io, while cuda‑oxide remains in early alpha and requires a pinned nightly toolchain with custom LLVM support. NVIDIA plans cross‑language interoperability among CUDA Rust, C++, and Python, allowing developers to choose frontends without lock‑in into a single ecosystem.
Related
Source: NVIDIA Developer | 2026-09-08