Home / BeaverDeck / Docs / Insights Guide / GPU Insights / GPU Idle Allocation
GPU Idle Allocation
BeaverDeck uses this check to identify a specific gpu condition that may need operator review.
| Check type | gpu-node-idle-allocation |
|---|---|
| Insights section | GPU Insights |
| Alert severity | Warning |
When It Reports A Finding
A Node advertises GPU capacity, but no active pod in the selected namespaces requests a GPU on that node.
Why This Is A Problem
An unused GPU node can represent significant cost and may indicate failed placement, excess capacity, or node pool scaling that is not following demand.
Recommended Response
- Confirm that all relevant namespaces are selected and that GPU workloads declare GPU requests.
- Review taints, tolerations, affinity, node selectors, and pending pod events for placement blockers.
- Consolidate or scale down the GPU node pool when the idle capacity is not intentional.
Scope And Limitations
The check uses scheduling requests, not DCGM utilization, and only sees selected namespaces. It can report idle when GPU workloads outside the selection are using the node.
After remediation: refresh GPU Insights and verify the underlying
resource or metric. Suppress the finding only when the condition is intentional and its risk is accepted.