This is a simple console app to monitor Nvidia GPUs. It is built with the purpose of helping ML engineers and hobbyists to help tune VRAM footprint as their app runs without having to string a series of print statements into their app or user the rather annoying alternative command line options.
Run this in a separate console window while you run your app. It will run continuosly until interrupted.
This is an extension on top of the nvsmi package which handles the integration work with the Nvidia SMI CLI.
Alternately with just nvidia-smi installed you can
nvidia-smi -l 2
but, it scrolls your console in a very annoying fashion, or
watch -n2.0 nvidia-smi
or in Powershell
while (1) {cls; nvidia-smi; sleep 2}
which may be too much information to fit on screen with multiple GPUs and is a bit noisy. Your call.
pip install nvsmi
(safe on Stability Diffusion LDM environment)
- or for more isolation
conda env create -f environment.yaml
conda activate nvsmi
python run.py