NVIDIA DCGM

Overview

NVIDIA Data Center GPU Manager (DCGM) provides libraries and tools for observing, managing, validating and integrating NVIDIA data-center accelerators. The companion DCGM Exporter exposes GPU telemetry in Prometheus text format; its documented metrics include DCGM_FI_DEV_POWER_USAGE as a gauge for board power draw and DCGM_FI_DEV_TOTAL_ENERGY_CONSUMPTION as a counter for cumulative GPU energy. The exporter repository ships source, container and package builds, a Helm chart and dashboards, and NVIDIA documents installing it, configuring Prometheus, the command-line reference and the metric list. DCGM also publishes C, C++ and Go interfaces plus command-line tools such as dcgmi and nv-hostengine.

Editorial verdict

Fits GPU platform teams that already run Prometheus and want board power draw and cumulative energy counters from NVIDIA data-center accelerators, with a Helm chart, dashboards and documented metric names to scrape. The trade-off is scope and attribution: it supplies GPU telemetry from NVIDIA's own stack, so per-process or per-tenant attribution and any carbon conversion sit outside it, and it covers NVIDIA accelerators only. Health checks and diagnostics come through dcgmi. Teams with mixed-vendor fleets or needing emissions reporting should pair it with other instrumentation.

Written by the GreenAIX editors from the recorded fields above. Vendors do not review or approve verdicts.

Use cases

• Monitor GPU power usage across data center nodes
• Collect cumulative GPU energy consumption counters
• Send GPU telemetry to Prometheus and Grafana
• Run GPU health checks and diagnostics with dcgmi

Sources

• DCGM Exporter GitHub repository
• DCGM Exporter README
• NVIDIA DCGM documentation
• DCGM Exporter metrics reference
• DCGM Exporter Apache 2.0 licence