GPU Monitoring

GPU Monitoring

Beszel can monitor GPU usage, temperature, and power draw for select devices. This feature is currently only available in the binary agent, not in the Docker agent.

Binary Agent Only

The Docker agent does not support GPU monitoring. You must use the binary agent if you need GPU metrics. For installation and usage details, refer to the official Beszel documentation on installing the binary agent.

AMD GPUs

Beszel uses rocm-smi to monitor AMD GPUs. Ensure rocm-smi is installed on the system running the agent.

• Installation Path: On Arch and Debian systems, installing rocm-smi-lib typically places the binary in /opt/rocm/bin.
• Symlink to /usr/local/bin: If /opt/rocm/bin isn't in the user's PATH, create a symlink:

  sudo ln -s /opt/rocm/bin/rocm-smi /usr/local/bin/rocm-smi

Nvidia GPUs

Beszel uses nvidia-smi to monitor Nvidia GPUs. This must be installed on the system.

• Nvidia Jetson Devices: Jetson boards are not compatible with nvidia-smi and are not currently supported. It may be possible to use tegrastats to track some metrics, but full support is not yet implemented.

Intel GPUs

Intel GPUs are not currently supported due to:

1. No test hardware is available to the developer.
2. There is no straightforward utility akin to nvidia-smi for real-time Intel GPU metrics (utilization, memory usage).

Please see issue #262 for more information or to track progress on Intel GPU support.