NVIDIA unveils CUDA Rust: cuda-oxide and cutile-rs, two tracks for writing GPU kernels
ssh4net · x · 2026-09-09
NVIDIA's technical blog details two tracks for writing GPU kernels in Rust:
- cuda-oxide: a custom rustc codegen backend compiling SIMT-style Rust kernels directly to PTX via the Pliron IR framework and LLVM; enforces memory safety at compile time with DisjointSlice and launch contracts; early alpha, requires pinned nightly toolchain and custom LLVM.
- cutile-rs: Tile-based GPU programming on stable Rust, with thread mapping and memory layout managed by the compiler through CUDA Tile IR JIT; already on crates.io and used in HuggingFace's Grout inference engine and mistral.rs; needs Rust 1.89+ and CUDA 13.3.
NVIDIA also plans inter-language interoperability between CUDA Rust, CUDA C++, and CUDA Python. The post includes getting-started commands and docs for both tracks.
Related event: NVIDIA announces CUDA Rust for writing GPU kernels in Rust(2 posts)→
More from coding & agent
- Fresh Memory, Stale Plans: PLANFENCE Blocks All Invalid Agent Actions Across 30 Workflows — rohanpaul_ai · 2026-09-09
- VoiceStudio open-sources a local ElevenLabs-style stack: voice clone, dubbing, 646 languages — thisguyknowsai · 2026-09-09
- Teknium says missing /review subagent is back — update to restore it — Teknium · 2026-09-09
- Layered summaries keep context at 8k, yet Qwen degrades after ~120 chat turns — Zeeplankton · 2026-09-09
- IndeHub Chapter 16: hands-on session on building, shipping and monetizing apps with AI agents — rudrank · 2026-09-09
- Users want AI assistants to talk to each other, but lock-in makes it unlikely — mobileraj · 2026-09-09