GPU

The GPU check monitors utilization, graphics memory usage, and temperature of a device's graphics cards. It runs locally on the device with the octoja agent on Windows and Linux (x64 and ARM64 each), every 5 minutes by default. macOS is not supported.

Steps

1. Open the device and choose Checks → Add Check → GPU.
2. Adjust which values are monitored and their thresholds if needed.
3. Click Add Check.

To assign the check to multiple devices, use a configuration package.

Settings

Utilization and memory thresholds are limited to 0–100 %, temperature thresholds to 0–150 °C. If a warning threshold is higher than its critical threshold, the check swaps the two values.

Data sources

On Windows, the check derives utilization from three samples taken one second apart and, like Task Manager, uses the busiest engine of the graphics card. The performance counters report no total graphics memory, so memory usage is not evaluated against the thresholds there.

Results

Each GPU is evaluated separately; the overall result is the most severe state across all GPUs. A value the data source does not provide is skipped.

• On Linux, an nvidia-smi error is reported as a failure; on Windows, the check falls back to the performance counters instead.
• With Ignore (OK), a device without a readable GPU reports OK. This does not confirm a healthy graphics card.

See also: Set up alerts for a check