NVIDIA DCGM

This snap includes NVIDIA DCGM and DCGM-Exporter to manage and monitor NVIDIA GPUs via the CLI or via Prometheus metrics.
Grafana dashboards can then be used to visualize the exported metrics, see for example:\ https://grafana.com/grafana/dashboards/12239-nvidia-dcgm-exporter-dashboard/

The snap includes the following components:
- DCGM: Data Center GPU Manager
- DCGM-Exporter: a Prometheus exporter for DCGM metrics

Please see the links at the bottom of the page for more details about the included components and their purpose.

How-To
---

How to install the snap:

sudo snap install dcgm

How to enable metrics collection:

# Start the DCGM-Exporter service (disabled by default)

sudo snap start dcgm.dcgm-exporter

# Get the metrics

curl -s localhost:9400/metrics

How to configure the snap services:

The NV-Hostengine and DCGM-Exporter services can be configured via the snap CLI.\
For example:

# Get all the configuration options

sudo snap get dcgm

# Set the NV-Hostengine port

sudo snap set dcgm nv-hostengine-port=5577

# Restart the NV-Hostengine service to apply the changes

sudo snap restart dcgm.nv-hostengine

Reference
---

Available configurations options:

- nv-hostengine-port: the port on which the NV-Hostengine listens.
The default is 5555.
- dcgm-exporter-address: the address DCGM-Exporter binds to.
The default is :9400.
- dcgm-exporter-metrics-file: the name of a custom CSV metrics file to be loaded by the exporter.
The path is assumed to be /var/snap/dcgm/common/.
The default metrics are located in /snap/dcgm/current/etc/dcgm-exporter/default-counters.csv.
Please refer to the DCGM-Exporter repository link at the bottom of the page for more information on the CSV file format.

Cryptography
---

During the snap build process, snapcraft downloads the CUDA keyring deb package using curl over HTTPS and verifies its integrity using SHA256 checksums.
The CUDA keyring deb package is then used to set up the appropriate source for the DCGM deb package, whose signature is verified using the keyring.\
For more information, see the CUDA keyring repository link and curl documentation at the bottom of the page.

Links
---
Upstream DCGM-Exporter repository\ https://github.com/NVIDIA/dcgm-exporter

Upstream DCGM repository\ https://github.com/NVIDIA/DCGM

DCGM Documentation\ https://docs.nvidia.com/datacenter/dcgm/latest/user-guide/index.html

Available NVIDIA GPU metrics\ https://docs.nvidia.com/datacenter/dcgm/latest/dcgm-api/dcgm-api-field-ids.html

Repository for the CUDA keyring and DCGM deb package\ https://developer.download.nvidia.com/compute/cuda/repos/

curl Documentation\ https://curl.se/docs/manpage.html