Installation

Prerequisites

This section describes the recommended dependencies to use nvCOMP.

• NVIDIA GPU with Volta architecture (sm70) or higher

• Linux distributions:

  • x86_64

    • Amazon Linux 2023

    • Azure Linux 3

    • Debian 12, 13

    • Fedora 44

    • KylinOS 11

    • OpenSUSE 15, 16

    • RHEL 8, 9, 10

    • Rocky Linux 8, 9, 10

    • SLES 15, 16

    • Ubuntu 22.04, 24.04, 26.04

  • arm64-sbsa

    • Amazon Linux 2023

    • Azure Linux 3

    • Debian 12

    • KylinOS 11

    • OpenSUSE 16

    • RHEL 8, 9, 10

    • SLES 15, 16

    • Ubuntu 22.04, 24.04, 26.04

• Microsoft Windows >= 10

• Microsoft Visual C++ 2015-2022 Redistributable >= 14

• NVIDIA driver >= 525.60.13 (Linux) and >= 527.41 (Windows)

• CUDA Toolkit >= 12.0

• Python >= 3.9

• GCC >= 8.5 (GCC >= 9.x on Linux x86_64)

• CMake >= 3.22

Installing the nvCOMP library

nvCOMP provides multiple ways to install the library. One can install the library through the package manager, through PyPI, or through the provided tarballs. The nvCOMP Developer Page provides the latest release of the library.

Installing the nvCOMP library through the package manager

nvCOMP offers distribution-specific packages for the library. Under each distribution, one can find the following packages:

• nvCOMP C/C++ for CUDA 12.x (development headers and libraries for CUDA 12.x)

  # Debian distributions (e.g., Ubuntu)
  libnvcomp5-dev-cuda-12
  
  # Red Hat distributions (e.g., RHEL)
  libnvcomp5-devel-cuda-12
  

• nvCOMP Runtime for CUDA 12.x (runtime libraries for CUDA 12.x)

  libnvcomp5-cuda-12
  

• nvCOMP Metapackage for CUDA 12.x (brings in the development headers and runtime libraries for CUDA 12.x)

  nvcomp-cuda-12
  

• nvCOMP C/C++ for CUDA 13.x (development headers and libraries for CUDA 13.x)

  # Debian distributions (e.g., Ubuntu)
  libnvcomp5-dev-cuda-13
  
  # Red Hat distributions (e.g., RHEL)
  libnvcomp5-devel-cuda-13
  

• nvCOMP Runtime for CUDA 13.x (runtime libraries for CUDA 13.x)

  libnvcomp5-cuda-13
  

• nvCOMP Metapackage for CUDA 13.x (brings in the development headers and runtime libraries for CUDA 13.x)

  nvcomp-cuda-13
  

Note that by installing the nvcomp-cuda-{12|13} metapackage, the libnvcomp5-{dev|devel}-cuda-{12|13} and libnvcomp5-cuda-{12|13} packages get installed automatically.

Further note that only the metapackages install symlinks in the default system directories (e.g., /usr/lib/x86_64-linux-gnu). The individual C/C++ and runtime packages do not install symlinks.

Installing the nvCOMP library through PyPI

nvCOMP offers complete Python integration for the library. One can use nvCOMP directly in Python by installing the nvidia-nvcomp-cu{12|13} module. The library also offers a standalone C/C++ module for nvCOMP. By installing the nvidia-libnvcomp-cu{12|13} module, only the development headers and runtime libraries are installed.

Note that by installing the nvidia-nvcomp-cu{12|13} module, the nvidia-libnvcomp-cu{12|13} module gets installed automatically.

• nvCOMP C/C++ for CUDA 12.x (development headers and runtime libraries for CUDA 12.x)

  pip install nvidia-libnvcomp-cu12
  

• nvCOMP Python for CUDA 12.x (complete Python integration for CUDA 12.x)

  pip install nvidia-nvcomp-cu12
  

• nvCOMP C/C++ for CUDA 13.x (development headers and runtime libraries for CUDA 13.x)

  pip install nvidia-libnvcomp-cu13
  

• nvCOMP Python for CUDA 13.x (complete Python integration for CUDA 13.x)

  pip install nvidia-nvcomp-cu13