Skip to content

CLI: probe --watch

Continuous GPU monitoring mode. Streams telemetry every 2 seconds until you press Ctrl+C. Perfect for watching GPU during inference or system stress tests.

What you'll see

Running hwledger probe --watch:

  • GPU memory usage updates every 2 seconds
  • Temperature readings refresh in real-time
  • Utilization percentage changes as workloads come and go
  • Clean Ctrl+C exit (<200 ms) with no leftover processes

This is useful for:

  • Watching a running inference job
  • Stress testing (check if GPU throttles under load)
  • Debugging OOM crashes (see memory spike)
  • Monitoring during fleet job execution

Journey not yet recorded.

Run the journey recorder to capture interactions:

./apps/macos/HwLedgerUITests/scripts/run-journeys.sh

What to watch for

  • Update interval: ~2s between refreshes (notice the timestamp change)
  • Memory changes: Watch as free VRAM drops when inference starts
  • Temperature: Rises gradually as GPU heats up under load
  • Utilization: 0% when idle, 85-100% during inference
  • Ctrl+C response: Press Ctrl+C and notice immediate clean exit (no hanging processes)

Next steps

Reproduce

bash
# Watch GPU for 30 seconds, then Ctrl+C
hwledger probe --watch

# JSON output for scripting
hwledger probe --watch --json | while read line; do
  echo "$line" | jq '.gpus[0].memory_free_gb'
done

Source

Recorded journey tape on GitHub

Released under the Apache 2.0 License.