Originally published at: https://developer.nvidia.com/blog/gpu-dashboards-in-jupyter-lab/
Learn how NVDashboard in Jupyter Lab is a great open-source package to monitor system resources for all GPU and RAPIDS users to achieve optimal performance and day to day model and workflow development.
The blog article was posted 3 days ago and contains long detailed “documentation” of how to install this extension using various way
But this article is already wrong (probably outdated) and misleading for potential users.
The article instructions to install:
pip install jupyterlab-nvdashboard
jupyter labextension install jupyterlab-nvdashboard
Gives:
ValueError: The extension "jupyterlab-nvdashboard" does not yet support the current version of JupyterLab.
According to the official docs (GitHub - rapidsai/jupyterlab-nvdashboard: A JupyterLab extension for displaying dashboards of GPU usage.) we only need to do the pip install (like for most extension on JupyterLab 3). And the extension is only supported on JupyterLab > 3, so the whole installation process documented in this 3 days-old article already does not exist anymore
@vincent.emonet Thank you for catching that. We have updated it.
It seems like it is still broken. In my case installing this package automatically updated jupyterlab
to v3.. and broke the Web UI. Uninstalling jupyterlab-nvdashboard
and downgrading jupyterlab
back to v2 solved the broken UI at least.
Are there any plans to update the dashboards to work with the current version of JupyterLab?
Could I also use this on VSCODE? Thanks!
I’m afraid not. Currently, only Jupyter Lab and the standalone dashboard is supported.
I recently tried setting up NVDashboard to monitor my GPU metrics in JupyterLab, and while the concept is fantastic, I ran into a few issues that made the process frustrating. First, after installing jupyterlab-nvdashboard
using pip install jupyterlab-nvdashboard
, the dashboards didn’t show up in JupyterLab as expected. I followed the instructions carefully, but the extension just wouldn’t load. I even tried reinstalling JupyterLab and the extension multiple times, but no luck.
Additionally, when I tried running the standalone Bokeh server using python -m jupyterlab_nvdashboard.server <port-number>
, I encountered errors related to missing dependencies. It seems like the documentation assumes all dependencies are pre-installed, but that wasn’t the case for me. I had to manually install Bokeh and PyNVML, which wasn’t mentioned in the setup guide.
Another issue I faced was with the JupyterLab extension not being compatible with my current JupyterLab version. I’m using JupyterLab 3.x, and the extension seems to have some compatibility issues. Has anyone else faced similar problems? If so, how did you resolve them?
On a side note, if you’re setting up JupyterLab from scratch, I found this guide really helpful: install jupyterlab on ubuntu. It walks you through the installation process step by step, which might save you some headaches.
If anyone has successfully set up NVDashboard and can share their experience or troubleshooting tips, I’d greatly appreciate it! Thanks in advance.