dcgmi

Show GPU discovery information

$ dcgmi discovery -l

$ dcgmi health -g [group_id]

$ dcgmi diag -r [1]

$ dcgmi dmon

$ dcgmi group -c [group_name]

$ dcgmi group -g [group_id] -a [gpu_id]

$ dcgmi topo -g [group_id]

dcgmi subsystem [options]

dcgmi is the command-line interface for NVIDIA's Data Center GPU Manager (DCGM). It provides monitoring, management, and diagnostic capabilities for NVIDIA GPUs in data center and HPC environments.
The tool enables administrators to monitor GPU health, run diagnostics, track performance metrics, and manage GPU groups for policy enforcement. It integrates with job schedulers and cluster management systems for automated GPU management.
DCGM tracks hundreds of GPU metrics including temperature, power, memory usage, and error counts. The diagnostic subsystem can detect hardware issues before they cause failures, supporting proactive maintenance.

SUBSYSTEM

Management subsystem: discovery, health, diag, dmon, group, topo, etc.

List discovered GPUs.

Check health of GPU group.

Run diagnostics (level 1-4).

Real-time monitoring dashboard.

Create named GPU group.

Show interconnect topology.

Display help information.

Requires NVIDIA DCGM service running on the host. Only works with supported NVIDIA data center GPUs. Some diagnostics require GPUs to be idle. Elevated privileges needed for certain operations.

DCGM was developed by NVIDIA and released around 2016 for enterprise GPU deployments. dcgmi provides CLI access to DCGM functionality, complementing the API and GUI interfaces for data center GPU fleet management.

nvidia-smi(1), nvtop(1), gpustat(1)