Last month CUDA-Oxide was introduced as an experimental Rust-to-CUDA compiler . From pure Rust programming language code, one can write CUDA GPU kernels in a "safe(ish)" manner with the CUDA-Oxide compiler emitting NVIDIA PTX output directly. Out today is the second update to CUDA-Oxide.

CUDA-Oxide 0.2 is the new version today and builds off their initial open-source code drop now with 37 pull requests from 23 contributors. This new version now makes it so a Rust GPU program is a single self-contained executable. There are also new capabilities added for making these Rust CUDA kernels more functional.

CUDA-Oxide 0.2 also brings improvements to ensure the kernels are more trustworthy, greater interoperability with CUDA Tile, and better upstream alignment with LLVM.

Downloads and more details on today's CUDA-Oxide 0.2 release from NVIDIA can be found at GitHub .