Certified Stacks
Certified Stacks are tested, versioned, one-command deployments of a vCluster plus a complete application stack. Each stack handles provisioning, dependency orchestration, and configuration automatically, so platform teams can go from zero to a fully configured multi-tenant environment without manual integration work.
Stacks are maintained in the certified-stacks repository. The repo contains full documentation, usage instructions, and architecture details for each stack.
Why use certified stacks
Deploying applications on top of Kubernetes involves stitching together multiple components — operators, controllers, CRDs, Helm charts — and getting the sequencing and configuration right across all of them. Certified Stacks eliminate that gap by providing pre-built, tested configurations that work out of the box.
- Multi-tenancy out of the box. Stacks ship with pre-built configurations for both hard and soft isolation models, so platform teams can match their tenancy architecture to their security and performance requirements without custom integration work.
- Built on validated reference architectures. Each stack extends partner-published reference architectures with additional deployment flexibility and tenancy options.
- Forkable and extensible. Every stack is a starting point, not a black box. Fork the repo, modify Helm values, swap components, or adapt the orchestration to your environment.
Tenancy models
Each stack supports two isolation models:
| Model | Description | Use case |
|---|---|---|
| Hard multitenancy | Each tenant gets a dedicated vCluster with private auto-provisioned nodes. Full isolation of control plane, GPU operators, drivers, and networking. | Untrusted tenants, strict compliance requirements |
| Soft multitenancy | Tenants share host cluster nodes via label-based assignment. Each tenant still gets its own vCluster with a separate Kubernetes API and workload scheduling. | Trusted tenants, cost-efficient shared infrastructure |
Available stacks
| Stack | Components |
|---|---|
| NVIDIA Run:ai | vCluster, NVIDIA GPU Operator, NVIDIA Run:ai |
| Slinky (Slurm) | vCluster, NVIDIA GPU Operator, Slurm via Slinky operator |
Get started
For detailed setup instructions, prerequisites, and configuration options, see the certified-stacks repository.
Build your own stack
Certified Stacks are designed to be customized. You can fork the repository and adapt any stack to your environment, or use the existing stacks as a reference for building your own. See the repository README for code style conventions and module structure.