Manage compute pools

This guide covers the day-to-day tasks for managing compute pools in Workbench: creating new pools, editing existing ones, deleting pools you no longer need, and assigning pools to sessions, training jobs, and endpoints.

Prerequisites

• At least one Kubernetes cluster with Workbench enabled.
• Admin or workspace-admin permissions on the cluster.

Create a compute pool

1. Navigate to Workbench > Compute Pools.

2. Click New Compute Pool in the top-right corner. A dialog opens.

3. Fill in the required fields:

   Field	Required	What it does
   Name	required	Unique identifier for the pool (e.g. workers-cpu-small).
   Min / Max Size	required	Autoscaling bounds. Min 0 enables scale-to-zero; max caps the number of nodes.
   Instance Type	required	The cloud machine type (e.g. g5.48xlarge). A type-ahead selector filters options by your cluster's provider and region.
   Workload	required	What this pool is optimized for: workspace, training, inference, cloud-shell, pvc-viewer, or remote-desktop.
   Root disk size (GB)	required	Boot disk size in GiB. Default is 30.
   GPU compute pool	optional	Enable the checkbox, then set GPU type, GPU count, and optionally GPU vendor.

4. Click Create.

The new pool appears in the list and is immediately available for selection when creating sessions, training jobs, or endpoints.

Edit a compute pool

1. Navigate to Workbench > Compute Pools.
2. Click the pool row to open its detail page.
3. Click Edit in the header. A dialog opens with the current values pre-filled.
4. Modify the fields you want to change. The Name field is read-only after creation.
5. Click Save.

note

Editing a pool does not affect resources that are already running on it. Changes apply only to new resources created after the update.

Delete a compute pool

1. Navigate to Workbench > Compute Pools.
2. Either:
   • Click the delete icon on the pool's row in the list, or
   • Open the pool's detail page and click the delete button in the header.
3. A confirmation dialog appears. Click Delete to confirm.

warning

Deleting a pool is irreversible. The autoscaler will drain any running nodes in the pool. Make sure no active sessions, training jobs, or endpoints depend on it before deleting.

Assign a compute pool to a session or job

You select a compute pool when creating a session, training job, or endpoint. The pool determines which instance type and GPU configuration back your workload.

Workload type	Where to select the pool
Session	Step 2 (Compute) of the session wizard — Compute pool dropdown. See Creating a session.
Training job	The compute section of the job form — Compute pool dropdown. See Submitting a job.
Endpoint	The compute section of the endpoint form — Compute pool dropdown. See Deploying an endpoint.

If you do not select a pool, the workspace's default pool is used.

View compute pool usage

1. Navigate to Workbench > Compute Pools.
2. Click a pool row to open its detail page.
3. The Used by card on the right shows a count of:
   • Sessions currently using this pool.
   • Endpoints currently using this pool.
   • Training jobs currently using this pool.

The Specification card shows the pool's GPU configuration, CPU, memory, storage, instance type, capacity, current availability, and estimated cost per hour.

Related

• Compute pools overview -- what pools are and how to choose the right shape.
• Reference -- every field on a compute pool.
• Creating a session -- how to launch a session on a pool.
• Submitting a training job -- how to run a job on a pool.