Faishme Memory Debug

Faishme Memory Debug is a stethoscope for your pipeline. It passes whatever value you feed it straight through untouched, and alongside it emits a text report of your system's memory usage at that exact moment - RAM used, VRAM used per GPU, plus a counter showing how many times it's run. If you're chasing an out-of-memory crash or just want to know which stage of your graph is the memory hog, you drop one of these in at each suspect point and compare the reports.

How it works

Under the hood it's two small calls: psutil.virtual_memory().used for system RAM and gpustat for per-GPU VRAM. Every time the node executes, it builds a string like:

Index: 3
System RAM used: 18.42 GB
GPU 0 VRAM used: 9.87 GB

The value input flows to value output unchanged - the debug info rides along as a separate debug_info STRING output. It's an output node, so it can sit at the end of a branch. The index counter increments every run, which is handy when the same node fires repeatedly in a batch and you want to know which iteration you're looking at.

Inputs and outputs

Only one input that matters:

• value - any type; it's just a passthrough so you can tap into the middle of a wire without breaking it.

Outputs:

• value - the same thing you put in.
• debug_info - the RAM/VRAM text report.

The dependency gotcha

This is the node that exposes the pack's sloppy requirements file. The code imports psutil and gpustat - gpustat is in requirements.txt, but psutil is not (and the file does list pandas and pyvips, which nothing in the pack uses). Most ComfyUI installs already have psutil floating around, but if this node throws at load time, that's the first thing to check:

cd ComfyUI/custom_nodes/ComfyUI_faishme
pip install psutil

Restart after. gpustat also needs nvidia-ml-py to actually read GPU stats on NVIDIA hardware - if it fails with an import error, that's the missing piece.

Installing the pack

Search "ComfyUI_faishme" in ComfyUI Manager, or:

cd ComfyUI/custom_nodes
git clone https://github.com/AkashKarnatak/ComfyUI_faishme

Restart ComfyUI. It shows up under FaishmeNodes.

Where it earns its keep

Honestly, this one is niche - the same info is available from nvidia-smi and the task manager, and the community's standard advice for VRAM pressure is a mix of --lowvram/offloading flags and lowering batch size rather than instrumenting the graph. Where this node wins is attribution: watching the counter climb inside a long batch run tells you which stage allocates and whether memory is being freed between iterations. If you're already this deep in a Faishme batch pipeline and something OOMs, it's the right tool. For everyone else, it's a curiosity that at least does exactly what it says.