Command Line#

This area documents customer-facing executables supplied by DCGM and closely related DCGM packages. Each command page provides exact command forms, accepted values, defaults, state lifetime, output, and failure behavior. Helper executables launched internally by another DCGM component are outside this scope.

These pages are comprehensive references rather than introductory guides. To learn DCGM concepts and operational workflows, start with Learn.

Available executables#

Customer-facing DCGM and DCGM Exporter executables#

Executable

Synopsis

Purpose

nv-hostengine

nv-hostengine [<options>]

Run the persistent DCGM host engine that serves dcgmi and DCGM API clients.

dcgmi

dcgmi <command> [<options>]

Discover, monitor, diagnose, and manage GPUs and other DCGM entities through a host engine. Its page links to each subcommand reference and documents their common conventions.

dcgmproftester

dcgmproftester<cuda-major> -t <field-list> [<options>]

Generate controlled CUDA workloads for DCGM profiling fields and optionally compare the generated activity with DCGM measurements. The CUDA major suffix is 11, 12, or 13.

dcgm-exporter

dcgm-exporter [<options>]

Select DCGM telemetry and expose it in Prometheus format, with optional entity and workload labels.