š Hardware Monitor¶
Built-in system resource monitoring that replaces Crystools ā no external custom node needed.
Overview¶
The Hardware Monitor displays CPU, RAM, GPU utilization, VRAM, and Temperature directly in the ComfyUI top bar. It also provides a contextual progress bar that shows what the pipeline is currently doing.
Settings¶
Available in Settings ā UmeAiRT ā Monitor:
| Setting | Default | Description |
|---|---|---|
| Enable Hardware Monitor | true |
Show/hide the monitoring bar. Disabling also stops the backend thread. |
| Monitor Style | Micro Gauges |
Visual style: Glassmorphism, Accent Strip, or Micro Gauges. |
| Refresh Rate | 1s |
Update frequency (0.5sā5s). Lower = more responsive but slightly more CPU usage. |
Styles¶
Micro Gauges (Default)¶
Compact rectangles with gradient fill, label at bottom-left, value at top-right.
Glassmorphism Pills¶
Frosted glass effect with backdrop blur, colored border, and thin fill bar at the bottom.
Accent Strip¶
Ultra-minimal ā colored label text with a 2px accent line. Best for small screens.
Progress Bar¶
During pipeline execution, a progress pill appears showing:
| Phase | Label Example |
|---|---|
| Image generation | Generating 45% |
| Upscaling (tile 1) | Upscaling 30% |
| Upscaling (tile 2+) | Upscaling T2 60% |
| Face detailing | Detailing 80% |
| Detail refining | Refining 55% |
| Video generation | Video Gen 20% |
| Frame interpolation | Interpolating 90% |
| Saving | Saving 100% |
Tip: Hover the progress pill for a detailed tooltip showing the current step, total steps, percentage, and internal node type.
After the pipeline finishes, the bar shows Idle after a brief 1.5s delay.
Platform Support¶
| Platform | GPU% | VRAM | Temperature | Library |
|---|---|---|---|---|
| NVIDIA (Windows/Linux) | ā | ā | ā | pynvml (auto-installed) |
| AMD ROCm (Linux) | ā | ā | ā | pyamdgpuinfo (manual install) |
| Apple Silicon (macOS) | ā | ā | ā | torch.mps (built-in) |
| CUDA Fallback | ā | ā | ā | torch.cuda (built-in) |
Multi-GPU setups (RunPod, etc.) show separate metrics per GPU.
On macOS, VRAM reports unified memory usage via torch.mps.driver_allocated_memory().
VRAM Tracking¶
- Peak VRAM: The monitor tracks peak VRAM usage per GPU.
- Reset: Double-click the VRAM pill to reset the peak counter.
- Tooltip: Hover to see current used/total, peak, and GPU name.
Technical Details¶
Backend (modules/monitor_hardware.py)¶
- GPUBackend abstract class with 4 implementations (cascade auto-detection)
- HardwareMonitor aggregates CPU, RAM, and GPU data
- MonitorService daemon thread broadcasts data via
PromptServer.send_sync('umeairt.monitor', data)
Frontend (web/umeairt_monitor.js + web/umeairt_monitor.css)¶
- ComfyUI extension registered via
app.registerExtension() - WebSocket listener for
umeairt.monitor(hardware data) - WebSocket listeners for
executingandprogress(pipeline progress) - Node type resolved via
app.graph.getNodeById()for contextual labels
API Routes¶
| Route | Method | Description |
|---|---|---|
/umeairt/monitor/gpu-info |
GET | Returns list of detected GPUs |
/umeairt/monitor/settings |
PATCH | Update rate and enabled state |