Pools
Compute pools answer one question for every other Workbench tab: where does this run? A pool bundles GPU type, GPU count per node, instance class, and autoscaling bounds — the same name then shows up in Sessions, Endpoints, Training Jobs, and Services.
The pools list
The list view shows every pool in your workspace. Each row displays:
- Pool name — the identifier.
- GPU — GPU type and count (or "CPU-only").
- Instance type — the cloud SKU.
- Size — maximum capacity (number of nodes).
- Available — current availability bar.
Click a pool to see its full specification, cost estimate, and which resources are currently using it.
Creating a pool
Click Create pool to define a new compute shape. You'll specify:
| Field | Required | What it does |
|---|---|---|
| Name | required | Unique identifier for the pool. |
| Min / Max size | required | Autoscaling bounds. Min 0 means scale-to-zero is allowed; max caps the blast radius. |
| Instance type | required | The cloud machine type (e.g. g5.48xlarge). |
| Workload | required | What this pool is optimized for (workspace, training, inference, etc.). |
| Root disk size | optional | Boot disk size in GiB. |
| GPU | optional | Enable GPU and select type and count. |
Picking the right pool
| Workload | Suggested pool shape |
|---|---|
| Notebook + light experimentation | 1x GPU (T4 / L4) or CPU-large; min=0 so it scales to zero when idle. |
| Single-node training run | 1x A100/H100 with whatever max your budget allows. |
| Distributed training | Multi-GPU pool (4x or 8x per node), node count set in the training-job sizing. |
| LLM inference | Pool sized to fit the model; H100x1 for 7-13B, H100x4 for 70B. |
| Heavy data prep / ETL | CPU-large with high memory; no GPU. |
info
Pools are managed by workspace admins. If you need a pool that doesn't exist, open an admin ticket with the use case and the rough hours/month estimate.
See Reference for every field on a compute pool.