Installing with Local Repositories

As an alternative to the method described in Creating a Local Mirror of the NVIDIA Repository, the DGX software can also be installed using local repositories. These repositories are provided by NVIDIA for installing the required packages from local repositories, allowing installation of packages without network access.

Installing the Base Red Hat Enterprise Linux System

The Red Hat components of the software must be installed either using the standard network based installation, or as described in Installing on “Air-Gapped” Systems. This includes registering and subscribing to the Red Hat Customer Portal.

If the system is connected to the network, follow the steps of the standard Red Hat Enterprise Linux installation, then enable the additional Red Hat repositories. Finally, execute:

sudo dnf update -y
sudo dnf install -y kernel-devel kernel-headers
sudo reboot

If the system is not connected to the network, follow the instructions in Creating a Local Mirror of the NVIDIA Repository

Downloading the Local Repositories

Download the packages that apply to your release.

Release 24.06

DGX Software local repository

dgx-local-repo-24.06-3.el9.x86_64.rpm

NVIDIA CUDA driver local repository

nvidia-driver-local-repo-rhel9-550.90.07-1.0-1.x86_64.rpm

nvidia-driver-local-repo-rhel9-535.183.01-1.0-1.x86_64.rpm

NVIDIA CUDA Toolkit local repository

cuda-repo-rhel9-12-4-local-12.4.1_550.54.15-1.x86_64.rpm

cuda-repo-rhel9-12-2-local-12.2.2_535.104.05-1.x86_64.rpm

Release 23.12

DGX Software local repository	dgx-local-repo-23.12-3.el9.x86_64.rpm
NVIDIA CUDA driver local repository	nvidia-driver-local-repo-rhel9-535.129.03-1.0-1.x86_64.rpm
NVIDIA CUDA Toolkit local repository	cuda-repo-rhel9-12-2-local-12.2.2_535.104.05-1.x86_64.rpm

Release 23.08

DGX Software local repository	dgx-local-repo-23.08-11.el9.x86_64.rpm
NVIDIA CUDA driver local repository	nvidia-driver-local-repo-rhel9-535.104.05-1.0-1.x86_64.rpm
NVIDIA CUDA Toolkit local repository	cuda-repo-rhel9-12-2-local-12.2.2_535.104.05-1.x86_64.rpm

Release 23.01

DGX Software local repository	dgx-local-repo-23.01-14.el9.x86_64.rpm
NVIDIA CUDA driver local repository	nvidia-driver-local-repo-rhel9-525.105.17-1.0-1.x86_64.rpm
NVIDIA CUDA Toolkit local repository	cuda-repo-rhel9-12-1-local-12.1.1_530.30.02-1.x86_64.rpm

Installing the Local Repositories

Install the local repositories:

sudo dnf install -y ./dgx-local-repo-24.06-3.el9.x86_64.rpm
sudo dnf install -y ./nvidia-driver-local-repo-rhel9-535.183.01-1.0-1.x86_64.rpm
sudo dnf install -y ./cuda-repo-rhel9-12-2-local-12.2.2_535.104.05-1.x86_64.rpm

With the local repositories installed, proceed with the standard Installing the DGX Software instructions, with the exception of the CUDA driver installation. To install the CUDA driver from the local repository, follow the steps in the next section Installing the NVIDIA CUDA Driver from the Local Repository.

Installing the NVIDIA CUDA Driver from the Local Repository

For non-NVSwitch systems, such as DGX-1, DGX Station, and DGX Station A100, install the driver using the default profile:

sudo dnf module install -y nvidia-driver:535-dkms
sudo dnf install -y --allowerasing nv-persistence-mode libnvidia-nscq-535

For NVSwitch systems, such as DGX-2, DGX A100, and DGX A800, install the driver using the fabric manager (fm) profile:
```
sudo dnf module install -y nvidia-driver:535-dkms/fm
sudo dnf install -y nv-persistence-mode nvidia-fm-enable
```

For DGX H100, install the DKMS version of the driver using the fabric manager (fm) profile:

sudo dnf module install -y nvidia-driver:535-dkms/fm
sudo dnf install -y nv-persistence-mode nvidia-fm-enable