The cuda-oxide SIMT track and the cutile-rs Tile track now allow the creation of GPU kernels using the Rust programming language. Nvidia released the cuda-oxide framework on a nightly basis, while the cutile-rs track is provided as a stable version for writing specialized GPU code.

This integration provides a method for developers to access the CUDA architecture without the requirement of C++ interfaces. By allowing for native kernels written in plain Rust, the new toolset supports the creation of safe, high-performance code designed specifically for Nvidia hardware.

Sign in to suggest edits

Key sources

  1. SOURCE@npashi“write CUDA kernels in plain Rust”x.com
  2. SUPPORT@rustaceans_rs“cuda-oxide (SIMT, nightly) and cutile-rs (Tile, stable)”x.com
Markdown