Installation

Rocky Linux from CIQ is deployed through pre-built images - GenericCloud (QCOW2) for on-premises virtualization, cloud marketplace images for public clouds, and bootc images for immutable container-native deployments. Existing Rocky Linux systems can also be converted to RLC using the Depot CLI.

System Requirements

Available Images

GenericCloud (QCOW2)

Pre-built virtual machine disk images for on-premises deployment using cloud-init for provisioning. Available for both RLC Pro and RLC+, in standard and NVIDIA (CUDA) variants.

GenericCloud Deployment Guide

Cloud Marketplace Images

Pre-configured RLC images available on major cloud providers:

Cloud images are available for both RLC Pro and RLC+ on x86_64 and aarch64 architectures. To find RLC images, go to portal.ciq.com, select your product, and you will be guided to the appropriate marketplace listing.

Bootc Images

Immutable, container-based operating system images using OCI container technology. Bootc images enable atomic updates, consistent deployments, and container-native OS management. Available from the CIQ Portal.

Bootc Images Guide

Conversion from Rocky Linux

Convert an existing Rocky Linux system to RLC Pro or RLC+ without reinstalling, using the Depot CLI.

Migration Guide

Image Specifications

All RLC images share these characteristics:

NVIDIA (CUDA) Images

NVIDIA images include the full CUDA toolkit and datacenter GPU drivers pre-installed:

• NVIDIA datacenter drivers (kmod-nvidia-dc-open)
• CUDA 12.8 SDK and runtime libraries (newer CUDA versions supported)
• cuDNN, NCCL, and supporting libraries
• nouveau kernel module blacklisted
• Larger disk footprint (30 GB minimum)

Post-Deployment Setup

RLC images ship with the Depot CLI pre-installed. After first boot:

1. Log in to Depot:

   sudo depot login -u [USER STRING] -t [USER TOKEN]
   

2. Enable your RLC product to activate package repositories:

   # For RLC Pro
   sudo depot enable rlc-pro-9
   
   # For RLC+
   sudo depot enable rlc-plus-9
   

3. Update the system:

   sudo dnf update -y
   

See the Getting Started guide for full details on Depot CLI setup.

Manually Installing GPU and Network Drivers

The NVIDIA (CUDA) images described above ship with GPU drivers pre-installed. On a standard image, or on a system converted with the Depot CLI, install the NVIDIA GPU and NVIDIA Networking (Mellanox) drivers from RLC Core. Complete the Depot enrollment above first.

Do not add NVIDIA's CUDA or DOCA repositories. RLC provides these packages itself: the drivers below come from RLC Core, and the CUDA Toolkit packages come from RLC Supplemental. Both repositories are activated by the depot enable step above, so there are no additional repositories to turn on.

Verify the Running Kernel

These drivers are kernel modules built for the CIQ kernel, and they do not load on any other kernel. Verify the running kernel before installing either driver:

uname -r

The version string must end in _ciq. Pre-built RLC images always boot a CIQ kernel. On a system converted with the Depot CLI, the conversion installs the CIQ kernel but leaves the previous kernel running, so a reboot is required before the suffix appears.

If a reboot does not produce a CIQ kernel, list what is installed:

rpm -q kernel

If a _ciq kernel is listed, it is not the default boot entry. Select it and reboot:

sudo grubby --set-default /boot/vmlinuz-$(rpm -q --qf "%{version}-%{release}.%{arch}\n" kernel | grep _ciq | sort -V | tail -1)

If no _ciq kernel is listed, confirm the RLC repositories are enabled, run sudo dnf update -y to install the CIQ kernel, then reboot.

NVIDIA GPU Drivers

Install the datacenter driver userspace and the matching open kernel modules:

sudo dnf -y install nvidia-dc-driver-latest kmod-nvidia-dc-open-latest

Reboot the system. After reconnecting, verify the driver:

nvidia-smi

nvidia-smi reports the GPU model, the driver version, and the supported CUDA version. If it cannot communicate with the driver, verify the running kernel as described above.

The -latest packages track the newest driver branch across upgrades. To hold a system on one branch instead, install the branch-specific packages, substituting the branch number for <branch>:

sudo dnf -y install nvidia-dc-driver<branch> kmod-nvidia-dc-open<branch> --allowerasing

Branch packages conflict with the -latest packages by design, so only one branch can be installed at a time. --allowerasing lets dnf remove the packages currently installed to complete the switch. List the available branches with dnf search nvidia-dc-driver.

NVIDIA Networking (Mellanox) Drivers

RLC Core provides CIQ-built Mellanox kernel modules and tooling:

sudo dnf -y install kmod-mlnx-ofa mlnx-tools

Reboot the system, then confirm the loaded module comes from the CIQ package rather than the in-tree driver:

rpm -qf "$(readlink -f "$(modinfo -n mlx5_core)")"

The result is one of the kmod-mlnx-ofa packages. A kernel package such as kernel-modules-core means the in-tree driver is still in use.

On a system that already has the NVIDIA GPU driver installed, this transaction also pulls in the NVIDIA peermem module, which provides GPUDirect RDMA between the GPU and the network adapter.

Kernel Updates

Driver modules are typically reused across a CIQ kernel update through weak-updates symlinks, so no rebuild is needed and the drivers load as before once the system reboots into the new kernel.

Where a new kernel is not compatible with the installed modules, the protected-kmods dnf plugin holds the kernel back rather than allowing an update that would leave the system without a working driver. The plugin is installed with the driver packages, which register themselves with it, so no configuration is required. A kernel held back this way is released once a matching driver build is published.

Confirm which module tree is in use at any time:

modinfo -n nvidia