JuliaGPU/CUDA.jl

CUDA programming in Julia.

What it solves

CUDA.jl provides a comprehensive programming interface for using NVIDIA CUDA GPUs within the Julia language, removing the need for users to manually install the CUDA Toolkit and simplifying the GPU-accelerated computing experience.

How it works

It acts as a meta-package that integrates several specialized subpackages:

  • CUDACore: Handles the main CUDA API wrappers and provides a compiler that allows developers to write CUDA kernels directly in Julia.
  • CUDATools: Provides developer utilities such as a profiler, NVML, and CUPTI.
  • Library Wrappers: Includes wrappers for essential NVIDIA libraries like cuBLAS, cuFFT, cuRAND, cuSPARSE, and cuSOLVER, with optional support for cuTENSOR, cuStateVec, cuTensorNet, and CUDNN.

The package automatically manages the CUDA Toolkit versioning, downloading a compatible version based on the NVIDIA driver and hardware available on the system.

Who it’s for

Developers and researchers using Julia who need to leverage NVIDIA GPUs for high-performance computing, custom kernel development, and linear algebra or signal processing tasks.

Highlights

  • Julia-native kernels: A compiler that allows writing CUDA kernels in Julia rather than C++.
  • Automated toolkit management: Automatically downloads and configures the suitable CUDA Toolkit version.
  • user-friendly array abstraction: Simplifies GPU memory and data management.
  • Extensive library support: Integrated wrappers for a wide range of NVIDIA's specialized mathematical and system libraries.

Related

  • Project
    kunzmi/managedCuda

    A .NET wrapper for NVIDIA's CUDA Driver API and libraries, enabling C# and Visual Basic developers to integrate GPU acceleration into their applications.

  • Project
    JuliaGPU/OpenCL.jl

    A Julia interface for the OpenCL parallel computation API that enables high-performance parallel programming on GPUs, FPGAs, DSPs, and multicore CPUs.

  • Project
    JuliaGPU/AMDGPU.jl

    A Julia package that enables programming and executing computations on AMD GPUs using the ROCm platform.

  • Project
    JuliaGPU/GPUArrays.jl

    A developer-focused toolkit for Julia that provides standardized interfaces and reusable functionality for creating new GPU array types across different hardware backends.

  • Project
    NVIDIA/cuda-python

    A collection of Python packages providing both high-level idiomatic and low-level binding access to NVIDIA's CUDA platform for GPU-accelerated computing.