Create a Work Cluster - Admin¶
Model inference and AI training workloads run on work clusters with GPUs. This tutorial creates a work cluster in Container Management and attaches GPU nodes, which is the foundation for deploying model services and agents.
Prerequisites¶
- Several nodes are ready, running the same OS and CPU architecture. It is recommended to use an OS supported by the installer.
- IPv4 forwarding is enabled on the nodes, and it is recommended to disable the firewall.
- Sign in to the console with a user who has the Container Management Module Admin role.
- The GPU nodes have the corresponding graphics drivers installed. During the node check, confirm that the GPU devices can be recognized.
Steps¶
-
In the left navigation bar, click Container Management -> Clusters, then click Create Cluster in the upper right corner.

-
Enter the basic cluster information and click Next:
- Cluster Name: lowercase letters, digits, and hyphens (
-), starting and ending with a lowercase letter or digit, up to 63 characters, e.g.drun-gpu-cluster. - Managed By: select the cluster that manages this cluster. The default is the global service cluster.
- Runtime:
containerdis recommended. - Kubernetes Version: choose a supported version; 1.29.5 is recommended.
- Cluster Name: lowercase letters, digits, and hyphens (
-
Enter the node configuration: fill in the IP and SSH credentials of the GPU nodes in the node list, then click Node Check. Continue after the check passes; if it fails, fix the node information as prompted and check again.
-
Adjust other parameters (network, storage, etc.) as needed, then click OK to create the cluster. The creation takes about 10–20 minutes.

-
After the cluster is created, check that the cluster status is Running in the cluster list and that the GPU nodes are ready in the node list.
Verify¶
Go to the cluster details page. If the GPU nodes show Healthy under Node Management, the cluster is ready to host AI workloads.
Next step¶
Continue to Deploy a Model Service.
To integrate an existing Kubernetes cluster instead of creating a new one, see Integrate Cluster. For more cluster and node management features, see Container Management.