Alauda GPU Management

DCGM-Exporter

Alauda Build of DCGM-Exporter is the NVIDIA GPU metrics component. It exposes GPU telemetry (utilization, memory, temperature, power, and health) so ACP monitoring and dashboards can collect it. In the unified delivery model it is one of the operands managed by Alauda GPU Management, and it is also referenced by name from HAMi deployments that monitor NVIDIA GPU backends.

TOC

Enable through Alauda GPU Management (recommended)Install as a standalone pluginVerifyUninstall

DCGM-Exporter is enabled by default in the ClusterPolicy instance. When Alauda GPU Management manages the GPU stack, no separate installation is required.

Confirm it is running on the GPU nodes:

kubectl get pods -l app=nvidia-dcgm-exporter -A

If it is disabled in your ClusterPolicy, enable the DCGM-Exporter component and reconcile the instance.

Install as a standalone plugin

Install DCGM-Exporter on its own when GPU metrics are needed on nodes that a HAMi backend exposes, or when Alauda GPU Management is not the metrics owner for those nodes.

  1. Go to Administrator > Marketplace > Cluster Plugin, switch to the target cluster, and deploy the Alauda Build of DCGM-Exporter cluster plugin.
  2. Set the node label so the exporter schedules onto the intended GPU nodes.

Verify

Confirm the exporter Pods are ready and metrics are exposed:

kubectl get pods -l app=nvidia-dcgm-exporter -A

Then continue to GPU dashboards. After the exporter has run for a short time, open Administrator > Operations Center > Monitor > Dashboards and switch to the GPU node and pod panels to see the metrics.

For monitoring as a usage scenario — enabling metrics for a GPU workload and handing off to ACP dashboards — see Enable GPU metrics.

Uninstall

To remove DCGM-Exporter, see Uninstall Alauda Build of DCGM-Exporter.