Ask HN: C++ still dominates on GPUs, why not Rust?

by Ambix on 6/6/2023, 9:40:21 PM with 2 comments
From what I know, modern GPUs are still programmed with C++ exclusively. See CUDA [0] for Nvidia and ROCm [1] for AMD.

Why is this? Why Rust is not loved there?

[0] https://docs.nvidia.com/cuda/

[1] https://github.com/ROCm-Developer-Tools/HIP

  • by thesuperbigfrog on 6/6/2023, 10:09:58 PM

    >> Why is this? Why Rust is not loved there?

    History and demand.

    CUDA is about 15 years old. MPI and similar parallel computing APIs are even older.

    C, C++, and Fortran have been the "goto" programming languages for parallel and scientific computing for decades.

    Rust is a relative newcomer that does not have enough demand yet to mandate CUDA APIs. It probably will some day.

    Another concern is the number of decades-mature libraries that are used in parallel computing domains. It seems much more likely that any Rust support will initially be Rust wrappers around C or C++ APIs rather than native Rust implementations.

    Some projects for Rust on CUDA are in progress:

    https://github.com/Rust-GPU/Rust-CUDA

    https://github.com/bheisler/RustaCUDA

  • by ActorNightly on 6/7/2023, 4:37:29 PM

    GPU programming =/= CPU programming. GPUs don't have complex execution paths with branching, and the programs that run are "memory safe" in the sense of what Rust aims to improve.