NVIDIA Enterprise RA Observability Guide
Type: Guide Tags: NVIDIA, Enterprise Reference Architecture, observability, telemetry, Grafana, Prometheus, DCGM, NetQ, NIM Operator, Kubernetes Related: NVIDIA-Enterprise-Reference-Architectures, NVIDIA-Enterprise-AI-Factory, NVIDIA-AI-Enterprise-Software-Reference-Architecture, NVIDIA-RTX-PRO-AI-Factory, NVIDIA-HGX-AI-Factory, NVIDIA-NVL72-AI-Factory, NVIDIA-DCGM, NVIDIA-GPU-Operator, NVIDIA-NIM-Operator, NVIDIA-NetQ, NVIDIA-Base-Command-Manager, NVIDIA-Mission-Control, NVIDIA-Run-ai Sources: https://docs.nvidia.com/enterprise-reference-architectures/index.html, https://docs.nvidia.com/enterprise-reference-architectures/observability-guide.pdf Last Updated: 2026-04-29
Summary
NVIDIA Enterprise RA Observability Guide is a production-oriented monitoring and alerting guide for NVIDIA Enterprise Reference Architecture environments. The current PDF focuses on AI factory observability using Kubernetes-native tooling, Prometheus, Grafana, DCGM, NIM Operator telemetry, BCM metrics, NetQ network metrics, dashboards, and alert templates.
Detail
Purpose
The guide gives Enterprise RA operators a repeatable way to monitor AI infrastructure from day one. It covers GPU, CPU, Kubernetes, network, and application telemetry so administrators can see health, utilization, performance, and workload impact across an AI factory.
Observability stack
- Prometheus and Grafana provide the time-series and dashboard layer.
- NVIDIA-DCGM and DCGM Exporter expose GPU health and utilization metrics.
- NVIDIA-GPU-Operator deploys GPU enablement and DCGM Exporter in Kubernetes environments.
- NVIDIA-NIM-Operator and NIM services provide application-side inference telemetry.
- NVIDIA-Base-Command-Manager can provide BCM metrics through a Prometheus data source.
- NVIDIA-NetQ provides network fabric telemetry and validation data.
- The guide includes dashboard and alerting templates for Kubernetes, GPU, CPU, disk, and network conditions.
Scope notes
The current guide explicitly focuses on dashboards and alerts. Logging and distributed tracing are marked as future/out-of-scope areas, so this wiki page should stay centered on metrics-driven observability until NVIDIA publishes a broader current guide.
NVIDIA context
Use this page for the cross-cutting observability guide across Enterprise RAs. Use NVIDIA-Mission-Control for integrated AI factory operations, NVIDIA-DCGM for GPU telemetry specifics, and NVIDIA-NetQ for network operations detail.
Connections
- NVIDIA-Enterprise-Reference-Architectures - program-level RA hub.
- NVIDIA-Enterprise-AI-Factory - strategic AI factory context.
- NVIDIA-AI-Enterprise-Software-Reference-Architecture - software stack that observability builds on.
- NVIDIA-RTX-PRO-AI-Factory, NVIDIA-HGX-AI-Factory, and NVIDIA-NVL72-AI-Factory - hardware RA targets for observability.
- NVIDIA-DCGM - GPU telemetry and health monitoring source.
- NVIDIA-GPU-Operator - Kubernetes deployment path for GPU software and DCGM Exporter.
- NVIDIA-NIM-Operator - application and model-serving layer relevant to observability.
- NVIDIA-NetQ - network telemetry and fabric validation source.
- NVIDIA-Base-Command-Manager - BCM metrics and cluster management integration.
- NVIDIA-Mission-Control - integrated operations plane for large DGX/NVL72 environments.
- NVIDIA-Run-ai - workload orchestration layer whose workloads need monitoring.
Source Excerpts
- NVIDIA’s Enterprise RA hub describes the Observability Guide as a standardized and production-ready reference for implementing observability in enterprise AI or HPC environments.
- The PDF states that the current guide focuses on advanced custom dashboard and alerting solutions for AI factories, using tools such as DCGM, NIM Operator, Prometheus Node Exporter, and other telemetry sources.