E2E CloudTIR Private Cluster
A TIR Private Cluster is a dedicated pool of GPU nodes that your account owns and your projects share. You create it yourself from the TIR portal, choose its GPU plan and number of nodes, and then hand its nodes to the projects that need them. Each project runs instances, model endpoints, training clusters, vector databases, and Slurm clusters on its nodes, and you move nodes between projects as their needs change, all within the same cluster.
Diagram as text
flowchart LR
C["Private cluster: 10 GPU nodes"] --> A["Project A: 4 nodes, training"]
C --> B["Project B: 3 nodes, inference"]
C --> F["3 nodes free"]Within the cluster, you can split each GPU into smaller instances with MIG, and steer workloads onto chosen nodes with node groups.
Who it is for
Section titled “Who it is for”A private cluster suits a team that runs long or predictable AI and ML jobs and needs GPUs it can count on. It also suits a team that wants to share one pool of GPUs across several projects, keep a bill that stays the same as usage changes, and control access by role.
| Private cluster | On-demand instance | |
|---|---|---|
| What you get | A pool of GPU nodes reserved for your account | One instance, from E2E’s available stock |
| Sharing across projects | Yes, each project takes a share of the nodes | One project only |
| What runs on it | Instances, endpoints, training, vector databases, and Slurm | The one instance |
| Best for | Long-running or shared work | Short-term or variable work |
For how the two are billed, see What is the difference between a private cluster and an on-demand instance?
Where to find it
Section titled “Where to find it”In TIR, open your project. In the side menu, under PRODUCTS, select Private Cluster. Your clusters are listed in Manage Private Cluster. Each cluster has these tabs: Project Allocations, Cluster Nodes, Cluster Details, Audit Logs, Node Monitoring, and Node Affinity.
What you can do
Section titled “What you can do”Create and share
Section titled “Create and share”- Create a private cluster with a GPU plan, a node count, and hourly or committed billing.
- Allocate nodes to projects, and take them back when a project is done with them.
Run work on it
Section titled “Run work on it”- Launch services on a private cluster: an instance, a model endpoint, a training cluster, a vector database, or a Slurm cluster.
- Use node affinity to keep a workload on, or off, a chosen set of the cluster’s nodes.
Monitor, grow, and delete
Section titled “Monitor, grow, and delete”- Monitor a private cluster: node health and usage, and the cluster’s audit log.
- Update a private cluster to rename it or add nodes to it.
- Delete a private cluster once you finish with it.
You can also list, create, and delete private clusters through the API; see Private Cluster in the TIR API.
Where to start
Section titled “Where to start”Start with Get started with TIR Private Cluster, which creates a cluster, gives its nodes to a project, and runs an instance on them. Nodes, projects, and access explains how nodes move between projects and who can do what. GPU partitioning with MIG covers splitting GPUs, and Billing options for TIR Private Cluster explains how hourly and committed clusters are charged. For help with a message, see Troubleshooting TIR Private Cluster, and for short answers, the TIR Private Cluster FAQ.