GPU Computing

GPUs are widely used in machine learning tasks including both training and inference, making them key components a performant machine learning ecosystem. Rust's GPU support, particularly around Geneneral-Purpose GPU (GPGPU) and compute shaders is incomplete, but a number of efforts including the WebGPU ecosystem, as well as CUDA and OpenCL are making progress.

wgpu [ crate · repo · docs ]

Cross-platform, safe, pure-rust graphics API

Last Commit: 2025-03-01 Last Published: 2025-01-23

vulkano [ crate · repo · docs ]

Safe wrapper for the Vulkan graphics API

Last Commit: 2025-02-28 Last Published: 2025-02-09

cudarc [ crate · repo · docs ]

Safe wrappers around CUDA apis

Last Commit: 2025-02-27 Last Published: 2025-02-27

blas [ crate · repo · docs ]

The package provides wrappers for BLAS (Fortran).

Last Commit: 2025-02-08 Last Published: 2025-02-08

cubecl [ crate · repo · docs ]

Multi-platform high-performance compute language extension for Rust.

Last Commit: 2025-02-28 Last Published: 2025-01-14

cubecl [ crate · repo · docs ]

Multi-platform high-performance compute language extension for Rust.

Last Commit: 2025-02-28 Last Published: 2025-01-14

ocl [ crate · repo · docs ]

OpenCL bindings and interfaces for Rust.

Last Commit: 2024-04-05 Last Published: 2024-03-27

cuda_std [ crate · repo · docs ]

Standard library for CUDA with rustc_codegen_nvvm

Last Commit: 2025-02-08 Last Published: 2022-02-07

dfdx [ crate · repo · docs ]

Ergonomic auto differentiation in Rust, with pytorch like apis.

Last Commit: 2024-07-23 Last Published: 2023-07-27

wonnx [ crate · repo · docs ]

Wonnx is an ONNX runtime based on wgpu aimed at being a universal GPU runtime, written in Rust.

Last Commit: 2024-07-21 Last Published: 2023-09-30

krnl [ crate · repo · docs ]

Safe, portable, high performance compute (GPGPU) kernels.

Last Commit: 2024-05-28 Last Published: 2024-05-28

arrayfire [ crate · repo · docs ]

ArrayFire is a high performance software library for parallel computing with an easy-to-use API. Its array based function set makes parallel programming simple. ArrayFire's multiple backends (CUDA, OpenCL and native CPU) make it platform independent and highly portable. A few lines of code in ArrayFire can replace dozens of lines of parallel computing code, saving you valuable time and lowering development costs. This crate provides Rust bindings for ArrayFire library.

Last Commit: 2023-09-24 Last Published: 2021-01-10