Global – November 10, 2026 -- CIQ says Rocky Linux from CIQ Pro AI (RLC Pro AI) deploys a GPU node in 3 minutes 44 seconds on the happy path and delivered 18% more model-loading throughput than Ubuntu 24.04 with the HWE kernel. The company presented the results at AI Infrastructure Field Day.
RLC Pro AI ships CUDA, NVIDIA drivers and tested frameworks inside the OS
The Enterprise Linux distribution bundles CUDA, the NVIDIA drivers, and frameworks such as PyTorch and Ollama, all tested together. On other distributions, CIQ says, a new GPU node means finding the right driver, compiling kernel modules, adjusting boot parameters and rebooting before a matching framework can be installed.
CIQ defines "validated" as specific versions of kernel, driver, CUDA toolkit and framework that are tuned, pinned, tested together and benchmarked for regressions. It describes RLC Pro AI as the first Enterprise Linux-compatible distribution authorized to ship the full NVIDIA AI stack.
CIQ Linux Kernel 6.18 posts 18% higher model-loading throughput than Ubuntu 24.04 HWE
The CIQ Linux Kernel (CLK) is built from the current upstream long-term release, 6.18, and CIQ patches and supports it. The 18% figure comes from a tuned kernel profile on a cold 8 GiB read from NVMe.
CIQ credits upstream scheduler improvements, not its own tuning, for faster NVMe reads against an older enterprise kernel. Some trading firms that employ kernel-tuning engineers have started requesting a supported kernel. They receive it as a repository package, with CVE and bug fixes delivered as repository updates.
Compile-time tuning falls outside Ansible, Puppet and Chef
CIQ argues that some parameters are set when the kernel, drivers and CUDA are compiled, so automation run after install cannot change them. The RLC Pro AI kernel profile tunes memory allocation and transparent huge pages for model loading and inference.
Golden images carry a separate cost. The kernel, drivers, CUDA and frameworks each release on their own schedules, forcing engineers to rebuild images repeatedly.
Four-tier container family keeps images and hosts on matched versions
The container family has four tiers: base, runtime, devel and workload. The devel image is free. Containers and host OS are versioned together, so a host driver upgrade does not require a container rebuild.
Containers run on the host kernel, so kernel gains come only from an RLC Pro AI host.
CIQ limits scope to the OS and urges workload-level testing
RLC Pro AI covers the operating system and validated frameworks; application tuning remains the customer's job. CIQ notes that in regulated environments some required security controls reduce GPU performance. Teams there run Rocky Linux from CIQ Pro Hardened on nodes that do not need peak GPU throughput.
A free developer license covers individuals on up to 25 nodes for non-production use.