As artificial intelligence workloads press harder against the limits of enterprise infrastructure, CIQ has updated its Rocky Linux-based AI distribution to the Linux 6.18 kernel — a quiet but consequential act of refusing the old bargain between stability and progress. For decades, enterprise computing has asked organizations to choose between keeping systems frozen and safe or chasing the performance that new hardware promises; RLC Pro AI is built on the premise that this is a false choice. In an era when the operating system beneath a GPU cluster quietly determines the return on enormous cap
CIQ Upgrades RLC Pro AI to Linux 6.18 for Faster GPU Inference
The OS layer decides how much you get from the GPU
Mark
So CIQ moved their AI operating system to a newer kernel. Why does that matter to anyone outside the data center?
Mimi
Because GPU inference is becoming how companies actually run AI in production. And the operating system kernel decides how much of the GPU's capacity the application can actually use. CIQ is saying they can give you the latest hardware support without sacrificing stability.
Mark
But don't all Linux distributions get the latest kernel eventually?
Mimi
They do, but it takes time. Traditional Enterprise Linux freezes the kernel at release and holds it for years. CIQ patches their kernel directly and ships updates between releases, so new GPU support reaches customers months faster.
Luke
How much faster, exactly? The announcement says "months faster" but doesn't give a number.
Mimi
That's fair. The concrete number they do give is the performance improvement: 33.8 percent faster model loading compared to the backported 5.14 kernel.
Mark
Is that a real-world number or a lab benchmark?
Luke
It's from CIQ's own testing. They validated the stack themselves. That's useful information, but it's one company's measurement under their conditions. Different workloads will see different gains.
Mimi
True, but the point isn't that every customer will see exactly 33.8 percent improvement. The point is that running a newer kernel with native driver support beats running an old kernel with backported drivers.
Mark
What about the licensing? CIQ charges per node, not per GPU?
Mimi
Right. So if you add more GPUs to a machine, your software bill doesn't go up. You only pay when you add another physical node.
Luke
That's a smart business model, but it also means CIQ is betting they can make money on support and services rather than per-accelerator licensing. That's a different bet than traditional Enterprise Linux vendors make.
Mimi
Exactly. CIQ is positioning themselves as the infrastructure layer for AI at scale, not as a general-purpose enterprise OS vendor.
O Pulso
- GPU-accelerated AI workloads are bottlenecked not by the hardware enterprises pay for, but by the aging kernels locked in place by traditional enterprise Linux distributions — a structural problem that compounds with every new accelerator generation.
- CIQ's update ships Linux 6.18, NVIDIA's 580 driver, and CUDA Toolkit as a single validated stack, delivering 33.8% faster model loading and a 1.38x improvement in time-to-first-token — gains that multiply across thousands of inference requests at scale.
- Rather than waiting years for a distribution release cycle, CIQ maintains its own kernel branch tracking upstream 6.18, pushing CVE fixes, Secure Boot-signed modules, and new accelerator support to customers months ahead of traditional enterprise Linux.
- A unified container image cadence means a configuration validated on a researcher's workstation can move — unchanged — to bare metal, cloud providers, or Kubernetes, collapsing the gap between experimentation and production.
- With per-node licensing rather than per-GPU pricing and direct upstream kernel contributions, CIQ is positioning itself as the accountable infrastructure layer for enterprise AI — one company responsible for whether the throughput organizations paid for actually arrives.
As artificial intelligence workloads press harder against the limits of enterprise infrastructure, CIQ has updated its Rocky Linux-based AI distribution to the Linux 6.18 kernel — a quiet but consequential act of refusing the old bargain between stability and progress. For decades, enterprise computing has asked organizations to choose between keeping systems frozen and safe or chasing the performance that new hardware promises; RLC Pro AI is built on the premise that this is a false choice. In an era when the operating system beneath a GPU cluster quietly determines the return on enormous capital investments, the question of who controls that layer — and how quickly they can move — has become a strategic one.
CIQ, the company behind Rocky Linux, has updated its AI-focused operating system to Linux 6.18, the latest long-term kernel release. The move targets a structural problem in enterprise computing: GPU-accelerated workloads run inside containers for portability, but the OS kernel beneath those containers determines how much of the GPU's capacity actually reaches the application. Traditional enterprise Linux distributions freeze their kernels at release and hold them for years, while GPU drivers and AI frameworks evolve rapidly. Teams have long faced a forced choice between stability and performance.
RLC Pro AI dissolves that choice. The system ships with Linux 6.18, NVIDIA's 580 driver, and CUDA Toolkit — all validated together at first boot. CIQ's measurements show model loading running 33.8 percent faster than the backported 5.14 kernel found in conventional enterprise Linux, and time-to-first-token improving by a factor of 1.38 under CPU contention. At inference scale, those gains compound across thousands of requests. CIQ maintains its own kernel branch tracking the upstream 6.18 series, allowing it to deliver new accelerator support and security patches months faster than traditional distribution release cycles.
Deployment flexibility is central to the design. Base, runtime, and development container images ship on the same cadence as the OS itself. A researcher can pull a free development image, serve a model locally with Ollama without assembling a CUDA environment, and then carry that exact validated configuration to bare metal, virtual machines, AWS, Azure, Google Cloud, or Kubernetes. The licensing model charges per node rather than per GPU, avoiding penalties for adding accelerators. CIQ engineers contribute fixes directly to the upstream Linux kernel, so defects surface at the source rather than accumulating in a private fork.
With more than 2.75 million Rocky Linux instances running worldwide and CIQ-supported technologies present in 90 percent of Fortune 100 companies, the 6.18 update signals a deliberate repositioning: not as a traditional enterprise Linux vendor, but as the infrastructure layer for AI at scale — the software that decides whether the return on GPU investment is actually realized.
CIQ, the company behind Rocky Linux, has moved its AI-focused operating system to Linux 6.18, the latest long-term kernel release. The shift addresses a persistent tension in enterprise computing: the need for stability in production systems versus the need for speed as new hardware and software arrive constantly.
The problem is structural. When enterprises deploy GPU-accelerated workloads—increasingly common across industries—those workloads run in containers for portability. But containers sit on top of an operating system, and the OS kernel decides how much of the GPU's capacity actually reaches the application. Traditional enterprise Linux distributions freeze their kernels at release and hold them steady for years. Meanwhile, GPU drivers, AI frameworks, and accelerator hardware evolve rapidly. Teams face a choice: lock in stability or chase performance.
RLC Pro AI breaks that choice by running the latest long-term kernel while maintaining the support and predictability enterprises expect. The system ships with Linux 6.18, NVIDIA's 580 driver, and CUDA Toolkit, all validated together at first boot. According to CIQ's measurements, model loading executes 33.8 percent faster than the backported 5.14 kernel found in traditional Enterprise Linux. Time to first token—a critical metric for inference—improves by a factor of 1.38 under CPU contention. For organizations running inference at scale, these gains compound across thousands of requests.
The technical mechanism matters. CIQ maintains its own Linux Kernel, or CLK, which tracks the upstream 6.18 series. Rather than waiting for the next distribution release, CIQ patches the kernel directly and ships updates to customers months faster than traditional Enterprise Linux upstream. This means support for new accelerators reaches production systems sooner. Brian Dawson, CIQ's Director of Product Management, frames the stakes plainly: teams spend enormous effort tuning frameworks, GPU kernel code, and models, only to have the OS layer beneath them become the bottleneck. The operating system decides whether they get the throughput they paid for.
Deployment flexibility is built in. Base, runtime, and development container images ship on the same cadence as the operating system itself. A researcher can pull the free development image from Quay.io or Docker Hub, run Ollama to serve a model locally without assembling a CUDA environment, and then move that exact same configuration to a workstation, to bare metal, to virtual machines, across AWS, Microsoft Azure, and Google Cloud, or into Kubernetes. One validated stack carries from experimentation into production.
The licensing model reflects CIQ's positioning. Rather than charging per GPU—which would penalize enterprises for adding accelerators—CIQ licenses per node. The company also handles security directly: RLC Pro AI ships with every CVE fix from its upstream long-term branch, Secure Boot-signed kernels and modules, and multi-year support from engineers with decades in HPC and Linux. Bjorn Hovland, CIQ's President, argues that enterprises commit enormous budgets to GPU infrastructure, and the software underneath decides the return. By building and supporting that layer themselves, CIQ makes one company accountable for the outcome.
CIQ engineers contribute directly to the upstream Linux kernel community, so defects customers surface get fixed at the source rather than remaining isolated to one vendor's fork. RLC Pro AI is available through CIQ's portal and through AWS, Microsoft Azure, and Google Cloud marketplaces, as installer ISOs, cloud images, and OCI container images. It joins CIQ's other Pro offerings—RLC+, RLC Pro, and RLC Pro Hardened—each optimized for different workloads and compliance requirements. More than 2.75 million Rocky Linux instances run worldwide, and 90 percent of Fortune 100 companies use CIQ-supported technologies. The move to 6.18 signals how the company is positioning itself: not as a traditional enterprise Linux vendor, but as the infrastructure layer for AI at scale.
Citações Notáveis
The OS layer decides how much of the accelerator capacity an enterprise paid for reaches the workload.— Brian Dawson, Director of Product Management at CIQ
Enterprises are committing enormous budgets to GPU infrastructure, and the software underneath decides how much of it they get back.— Bjorn Hovland, President of CIQ
Quer a matéria completa?
Leia o original em HPCwire ↗
Technology