Diagnostics

This section defines the public reference contracts for DCGM’s single-node diagnostic module. Use it to determine how diagnostic configuration is resolved, what each available plugin tests, and how to interpret reported errors. For task-oriented guidance about choosing and preparing a diagnostic run, see Diagnostics.

The dcgmi diag reference remains authoritative for command syntax, suite and test selection, target entities, common options, output, and process exit status.

Contents

Diagnostic reference sections

Section	Use it to
Diagnostic Configuration File	Define dcgm-diag-v1 SKU overlays and understand GPU matching, recursive merging, source precedence, and common configuration fields.
Diagnostic Errors	Interpret DCGM_FR_* identifiers, categories, severities, and the response guidance carried by diagnostic results.
Diagnostic Plugins	Choose a diagnostic test and look up its purpose, prerequisites, parameters, execution behavior, failure signals, and result semantics.

How the references fit together

A diagnostic request selects a predefined suite or one or more canonical test names. At runtime, the diagnostic module determines which plugins are available for the selected entities and resolves each plugin’s effective parameters from its fallback values, the packaged SKU configuration, an optional configuration-file overlay, and command-line overrides.

The selected plugins then return per-test and per-entity statuses, messages, and, when applicable, diagnostic error codes. A Pass, Fail, or Skip status describes the outcome of that test invocation; the associated error code and message distinguish a diagnostic finding from a setup, execution, or availability problem. Consult both the applicable plugin page and the error reference when interpreting a result.

These pages cover local active diagnostics. Coordinated multi-node diagnostics use a separate module, command, configuration model, and diagnostic catalogue; see Multi-Node Diagnostics.