CIQ partner
Rocky Linux for AI: GPUs ready from first boot
With RLC Pro AI, the NVIDIA driver, CUDA and PyTorch arrive installed and tested against the kernel. We deploy it on your GPU servers and look after it, so your team starts on the models on day one.
Licensed per server, not per GPU. For x86_64 and ARM.
What it takes to keep a GPU running on Linux
On community Rocky Linux, a working GPU depends on the NVIDIA driver, the kernel modules, CUDA and PyTorch all fitting together. A kernel update can leave the server without its GPU until someone rebuilds the modules, and every new machine costs another half hour or more. With ten servers, that is work every single week.
RLC Pro AI removes that part. CIQ tests the whole stack with its own kernel and ships it together, so an update does not break the combination, and we take care of rolling it out.
| Kernel | CIQ Linux Kernel, tracking kernel.org long-term releases and tuned for model loading. |
|---|---|
| NVIDIA | Data centre driver 580 with open kernel modules, CUDA 12.8, cuDNN, NCCL and Fabric Manager. |
| Frameworks | PyTorch 2.8 and TorchVision, built by CIQ with CUDA enabled from first boot. |
| Foundation | Everything in RLC Pro: Rocky Linux 9, CIQ patches and legal indemnification. Supported until May 2032. |
| Formats | ISO, images for AWS, Azure and Google Cloud, OCI container and bootc, each with its SBOM. |
| Licence | Per server, not per GPU, on x86_64 and ARM. |
How much faster than Ubuntu
The gap comes from the tuned kernel and from not having to adapt the stack by hand, which is exactly the part that tends to be left half done.
RLC Pro AI vs Ubuntu*same hardware
Is RLC Pro AI what you need?
You run NVIDIA GPUs and want the stack ready
Servers that train or serve models, with the driver, CUDA and PyTorch already matched and someone behind it when something breaks.
No GPU: RLC Pro
If what matters is staying on one release for years.
The whole platform
Storage, Kubernetes and the models running, built by us.
Just trying it
CIQ offers a free developer licence for up to 25 servers, outside production.
Choosing hardware? See which server you need for a local LLM. All editions on Rocky Linux support.
What we bring
CIQ prepares the system; choosing the hardware, deploying it and getting the models running is on us.
- Before
Which server and which GPU
Based on the models you want to run, with Dell or Supermicro servers if needed.
- Subscription
Only what you need
RLC Pro AI per server, mixed with RLC Pro where there is no GPU.
- Rollout
The same on every node
From CIQ's ISO or images, with Ansible, in your data centre or in the cloud.
- Models
From GPU to first model
Ollama, vLLM or your own containers on top, with the open models you choose.
- Day to day
Support in English, Spanish or French
24/7 or 8×5, with CIQ behind us for system and NVIDIA stack issues.
Frequently asked questions
What is RLC Pro AI?
CIQ's edition of Rocky Linux for servers that run AI. It builds on RLC Pro and adds the CIQ Linux Kernel, the NVIDIA data centre driver, CUDA, cuDNN, NCCL and PyTorch, installed and tested together.
What do you do and what does CIQ do?
CIQ publishes the operating system and its patches. We help you choose the hardware, deploy it, put the models on top and are the people you call when something breaks. If the problem is in the OS or the NVIDIA stack, we escalate it to CIQ ourselves.
Can't I just install the NVIDIA driver myself?
Yes, on community Rocky Linux you can, following NVIDIA's guide. CIQ puts it at 30 to 60 minutes per server, and the real cost comes later: rebuilding the modules for every new kernel and checking that CUDA and PyTorch still match the driver, on every machine. RLC Pro AI takes that off your plate.
Which versions does it ship?
The current release is based on Rocky Linux 9.7, with the NVIDIA 580 driver (open kernel modules), CUDA 12.8 and PyTorch 2.8. It is supported until 31 May 2032.
Is it licensed per GPU?
No. It is licensed per server, so adding cards does not raise the subscription. To try it, CIQ offers a free developer licence for up to 25 servers, outside production.
Does it work without a GPU or on ARM?
Yes. The GPU is optional and there are builds for x86_64 and aarch64. PyTorch picks up the NVIDIA GPU automatically when it finds one.
Tell us what you want to run
With the models, the number of users and the hardware you have, we will tell you what to build and on how many servers.
