Cluster Templates Overview
Overview
When you create a Kubernetes cluster in Armada Bridge, you choose a cluster template. The template determines what software and capabilities are provisioned on the cluster (e.g., a basic Kubernetes cluster, Ray for distributed workloads, JupyterHub with KAI Scheduler, or NVIDIA NIM for inference).
As a Tenant Admin, you create and deploy Kubernetes clusters on top of the infrastructure (Bare Metal or Virtual Machine) allocated to your tenant. Select the template that matches your workload.
This section describes the cluster templates supported by Bridge and links to the steps for creating each type.
Supported Cluster Templates
| Template | Use case | Documentation |
|---|---|---|
| None (Basic) | Plain Kubernetes cluster with no additional stack. Use for custom workloads or when you will install your own tools. | None (Basic) |
| Ray-enabled | Kubernetes cluster with KubeRay for distributed computing and Ray workloads (e.g., model serving, data processing). | Ray-Enabled |
| JupyterHub with KAI Scheduler | JupyterHub cluster with KAI Scheduler for interactive notebooks and job scheduling. Use for data science and ML development. | JupyterHub with KAI Scheduler |
| NVIDIA NIM | Kubernetes cluster with NVIDIA NIM for high-performance, GPU-accelerated inference. Use for deploying and serving AI/LLM models. | NVIDIA NIM |
Choosing a Template
- Basic Kubernetes — Choose None (Basic) when you need a standard cluster and will add your own components.
- Model serving / Ray workloads — Choose Ray-enabled when you need KubeRay for model deployment or Ray-based applications (e.g., LLM endpoints).
- Interactive notebooks — Choose JupyterHub with KAI Scheduler when you need JupyterHub for notebooks and the KAI scheduler.
- GPU inference (NIM) — Choose NVIDIA NIM when you need optimized inference microservices for AI/LLM models.
Prerequisites
- Tenant Admin access — Log in as a Tenant Admin to create clusters.
- Compute resources — Bare Metal or Virtual Machine resources allocated to your tenant.