Muse Run Stats
A frontend stopwatch for the whole ComfyUI workflow that also tracks average GPU temperature, GPU utilization, VRAM, and system RAM while it runs. Starts counting on Run, shows total time plus the averages collected along the way on finish. Doesn't need to be wired to anything.
Muse Run Stats
A live stopwatch for a ComfyUI workflow, with GPU temperature, GPU utilization, VRAM, and system RAM tracked alongside it.
Built by Muse Collective — a single node you drop anywhere on the canvas. It doesn't need to be wired to anything and never appears in the executed prompt. From the moment you hit Run, it counts elapsed time and polls your machine's real hardware stats every ~1.5 seconds, so on a long generation (MiniMax H3 renders especially) you can see at a glance whether it's genuinely still working or has stalled — a run stuck at 100% GPU with VRAM pegged and no progress looks very different from one that's simply mid-step.
What it shows
- Elapsed time — starts the moment a prompt begins executing, stops when it finishes, errors, or is interrupted.
- GPU temperature (°C) and GPU utilization (%) — via
nvidia-smi. - VRAM used / total — via
nvidia-smi. - System RAM used / total — via
psutil.
Every value falls back gracefully if it can't be read (no NVIDIA GPU, nvidia-smi not on PATH, psutil not installed) — you still get whatever readings are available rather than the node breaking or showing nothing at all.
Installation
Via ComfyUI Manager
Search for Muse Run Stats and click Install.
Manual
cd ComfyUI/custom_nodes
git clone https://github.com/muse-collective-26/Muse-Run-Stats
Restart ComfyUI after installing.
Python packages
pip install psutil
psutil is used for the system RAM reading. GPU stats use nvidia-smi directly (assumed already on PATH with any working NVIDIA driver install) — no extra package needed for that half.
Usage
- Add a Muse Run Stats node anywhere on your canvas.
- Run your workflow as normal — no wiring needed.
- Watch the node update live: elapsed time and current hardware readings while running, a final summary (total time + the averages collected along the way) when it finishes.
Why nvidia-smi instead of pynvml
nvidia-smi's CSV output is simple enough to shell out to per poll, cheaply and with zero extra Python dependency. Only the first GPU is read — this node is aimed at the common single-card workflow, not multi-GPU averaging.
Credits & Licensing
This repository's code is licensed under the MIT License — see LICENSE. It has no model or third-party weight dependency of its own; it only reads hardware stats already exposed by your GPU driver (nvidia-smi) and OS (psutil).