Skip to main content

GPU Dashboard

The GPU Dashboard provides a Kubernetes perspective to track relationships between Node - GPU (MIG) - Pod.

  • Visually maps Node - GPU (MIG) - Pod relationships to easily understand GPU resource allocation.
  • Displays Top 5 trends by utilization, temperature, and memory to quickly detect overuse or imbalance of resources.
  • Highlights critical states such as Pending or Unused GPUs to identify allocation gaps or abnormal usage patterns at a glance.

Permissions & Requirements​

  • Supported environment: Kubernetes cluster project
  • Agent version: Kubernetes Agent v1.8.7 or higher
  • Requires Open Agent installation

Main Screen​

A visual dashboard to easily identify GPU resource status and usage within the cluster.

GPU dashboard

GPU Resource Summary​

Summarized GPU information (assigned nodes, Pods, GPU counts by status) collected during the last 5 minutes, shown in four widgets.

GPU Map​

Displays collected devices at the query time in a map chart.

  • Physical devices are labeled P, MIG instances are labeled M.
  • Grouping can be done by Node/Physical device, with options to color by status or utilization.

Usage​

Shows the total cluster VRAM size and usage, average GPU utilization per device, and VRAM usage over the last 1 minute.

GPU Performance Summary (Top 5)​

Displays performance trends of major physical device metrics during the query period.

  • Utilization (%)
  • VRAM Usage (MiB)
  • Temperature (℃)
  • SM Active (%)

GPU / Node / Pod Lists​

Lists of GPUs, Nodes, and Pods are displayed.

  • Node and Pod lists show the Top 5 items by GPU utilization.
  • The GPU list shows all GPUs collected at query time (data collected within the last 1 minute).

GPU top5 list

Details​

Click the details icon next to a GPU in the GPU Map or GPU List to view the relationship map and metric trends for the selected GPU.

GPU dashboard details

Last updated on