GPU Feature

This guide provides information about the agent environment and GPU options required to enable GPU monitoring.

GPU Monitoring Support Environment

By default, performance and inventory data can be collected using nvidia-smi. However, if nvidia-dcgm is installed, additional metrics can be collected.
When nvidia-dcgm is present, the nvidia-dcgm service must be running.

• You can verify that the nvidia-dcgm service is running properly with the following command: systemctl status nvidia-dcgm

• The list of collectable metrics may continue to grow over time.

For details on metrics by collection tool, see GPU Metrics.

GPU Features

To use the GPU Inventory and GPU Performance Summary features, you must update the agent to the required version and apply the necessary settings.

Tip

Supported Version

Linux minimum 2.9.3, recommended 2.9.16 or later

• Amazon Linux, RHEL, Fedora, Oracle Linux, CentOS, Rocky Linux, SUSE

  sudo yum update whatap-infra

• Debian, Ubuntu

  sudo apt-get update
  sudo apt-get install whatap-infra

GPU Options

To enable the GPU Inventory feature, the following options must be enabled:

• nvidiav2.enabled Boolean

  Default false

  Option to collect NVIDIA GPU metrics including physical GPU configuration and resource usage.

• nvidiav2.mig.enabled Boolean

  Default false

  Option to collect metrics related to NVIDIA GPU MIG configuration and resource usage. Requires nvidiav2.enabled to be set to true.

• nvidiav2.proc.enabled Boolean

  Default false

  Option to collect information about processes assigned to NVIDIA GPUs. Requires nvidiav2.enabled to be set to true.

• nvidiav2.dcp.enabled Boolean

  Default true

  Option to collect NVIDIA DCP (Data Center Profiling) metrics. Requires nvidiav2.enabled to be set to true.

• nvidiav2.interval.sec

  Default 30 seconds

  Interval for collecting NVIDIA GPU metrics.