AMD and Rancher Government Solutions Bring Security-Focused HPC and AI to Kubernetes
Nov 16, 2026
AMD and RGS simplify HPC and AI scheduling on Kubernetes with a hardened architecture built for efficient GPU utilization.
AMD powers the world's leading scientific computing, from the exascale supercomputers driving discovery in the United States and Europe to Lux, the first Genesis Mission system for AI-driven science. With leadership CPUs, GPUs, FPGAs, networking and open ROCm software, AMD gives nations, laboratories and enterprises a single open platform for high-performance computing, AI for science, sovereign AI, agentic AI and the path to quantum computing.
AMD and Rancher® Government Solutions (RGS) have partnered to deliver a hardened, production-ready reference architecture designed to simplify HPC and AI workload scheduling and improve GPU utilization on Kubernetes. Powered by the RGS Carbide Platform and high-performance AMD Instinct™ GPUs, this solution empowers research and national security organizations with a hardened, highly scalable foundation for their most critical GPU-accelerated workloads. By integrating open-source projects Slinky and Kueue, organizations gain intelligent, dynamic job scheduling and automated GPU allocation to drive peak efficiency.
AMD brings a proven record of leadership in accelerated computing, powering exascale supercomputers, demanding scientific and national security workloads, and large-scale AI inference. AMD Instinct™ accelerators and open ROCm™ software provide a foundation for advancing HPC and AI across diverse deployment environments. This experience informs an open, repeatable reference architecture designed to help more organizations deploy and operate GPU-accelerated infrastructure with greater confidence.
“HPC and AI are converging quickly. As workloads grow, organizations need to share GPU resources efficiently while meeting high security standards.” says Chris Reynolds, VP of Solution Architects at RGS, “This reference architecture is built to do both. Because it runs on Kubernetes, researchers can develop on a single workstation and scale the same work to a large GPU cluster with a consistent baseline.”
From workstation to large-scale cluster, with one definition
Traditional HPC systems deliver enormous compute power, but getting work onto them is often slow. Software is installed directly on the shared system, its environment rarely matches a researcher's own machine, and a job that works on a laptop can fail on the cluster. Researchers end up debugging environments and waiting for time on the big system just to test small changes.
Containers solve the environment problem. Each job is packaged with its code, libraries, and dependencies, so it runs the same way everywhere. Nothing needs to be installed on the host, and one team's software can't conflict with another's.
Kubernetes as the container orchestration layer then solves the scale problem with a declarative model. Instead of writing step-by-step scripts for each system, a researcher describes what a job needs, such as the containers to run, how many GPUs, and how much memory. Kubernetes matches workloads to the right hardware, ensures high availability and resilience by distributing tasks across the cluster, and automatically reschedules jobs if a hardware failure occurs.
HPC job definitions are plain text files that can be version-controlled, reviewed, and shared. Experiments become reproducible, and jobs can be re-run months later in the same environment. Teams can work on pieces of a problem in parallel, and the large cluster is reserved for the runs that actually need it.
A single shared cluster can also run batch research jobs alongside long-running services, such as model inference, instead of splitting hardware into separate silos. This approach helps reduce idle capacity and improve utilization, so organizations can get more value from the hardware they already own.
The same declarative approach applies to the platform itself. AMD and RGS used it to automate deployment of the full stack, from GPU drivers to the job scheduler, so organizations can stand up identical environments repeatably. Data scientists and researchers can focus on their work instead of building and maintaining the infrastructure underneath it.
A hardened foundation for sensitive workloads
HPC environments often handle sensitive and critical data, so a secure, compliant baseline is essential. Running AI and HPC efficiently also takes close integration between hardware and software.
AMD and RGS validated the core components of this architecture to confirm that the environment reliably uses AMD GPUs and matches each workload with the accelerators it needs. The foundation includes:
AMD Capabilities |
RGS Capabilities |
Optimized Compute: High-performance AMD Instinct GPUs with automated, declarative installation of the AMD ROCm software stack. |
Security-Focus Orchestration: RKE2 Government, a hardened Kubernetes distribution running on secure enterprise Linux designed to meet U.S. Government security requirements. |
GPU Management: AMD ROCm GPU Operator to automatically expose and manage GPU resources for containerized workloads. |
Hardened Baseline & Automation: Carbide Platform is designed to provide a secure, hardened foundation with automated full-stack deployment from drivers to job scheduling. |
Fair, efficient GPU scheduling with Slinky and Kueue
A shared GPU cluster needs a clear way to decide which jobs run, when, and on what hardware. This architecture pairs two open-source projects to handle that challenge.
- Slinky brings Slurm, the scheduler many HPC users already know, to Kubernetes, so existing workflows carry over.
- Kueue manages job queues and quotas, so teams share GPU capacity fairly and resources don't sit idle.
Together, they let researchers run demanding applications reliably, including vLLM for high-throughput large language model inference.
Sovereign AI requires both powerful compute and operational control. AMD is a leader in enabling this foundation through its portfolio of high-performance processors, Instinct accelerators, networking, and open ROCm™ software. Building on the AMD, SUSE, and RGS collaboration announced at Advancing AI 2026, this reference architecture brings that vision to HPC and AI scheduling: combining AMD acceleration with RGS’s hardened Kubernetes foundation to help government agencies, research institutions, and regulated enterprises run sensitive workloads within environments they control.
Accelerating the Future of Scientific Computing: More to Come
This jointly developed reference architecture brings traditional HPC job scheduling and modern container orchestration together on a secure foundation.
Together, AMD and RGS are giving private enterprises, research institutions, and national laboratories the secure, open infrastructure they need to tackle complex scientific challenges, so their teams can spend less time on infrastructure and more time on discovery.
Related Blogs
Footnotes
AMD, the AMD Arrow logo, AMD Instinct, ROCm and combinations thereof are trademarks of Advanced Micro Devices, Inc. RGS is a trademark of SUSE LLC. Other product names used in this publication are for identification purposes only and may be trademarks of their respective owners.
AMD, the AMD Arrow logo, AMD Instinct, ROCm and combinations thereof are trademarks of Advanced Micro Devices, Inc. RGS is a trademark of SUSE LLC. Other product names used in this publication are for identification purposes only and may be trademarks of their respective owners.