Jupter_HPC
Overview
What is Jupyter and why is it useful?
Jupyter gives you a browser-based, multi-language suite for programming, data analysis and visualization with e.g. python and R. Jupyter notebooks are structured in consecutive “cells” (meaningful steps / sections). These may contain either code and the results of its evaluation, or supporting comments, texts and figures (usually in Markdown).
Code and/or parameters can be changed and the whole notebook or just individual cells can be re-evaluated at any time. This makes Jupyter a highly interactive tool for gaining quick insights into data, testing new code and visualization, and bundling data, code, results and supporting information.
Therefore Jupyter can be very useful for communicating your research and for collaborating with other persons. The Jupyter notebook format is also gaining popularity as a form of scientific output on its own.
Benefits of Jupyter_HPC
Our Jupyter_HPC service (JupyterHub) enables you to use the popular JupyterLab interface without bothering to install and maintain software on your local computer. You also avoid burdening your workstation with heavy computation by using our central resources instead.
Running Jupyter on MaRC3 also gives you access to all file systems, directories and software which you could access if you were using the MaRC3 cluster via command line or other applications. Thus, Jupyter_HPC provides an interactive, low-barrier interface to use our HPC infrastructure.
For code execution there is a python kernel available by default. Kernels for other languages like R can also be activated. You are free to install Jupyter add-ons, python packages, use virtual environments and the HPC module system.
Limitations and references
Jupyter_HPC is generally meant for testing, development, and exploratory data analysis. Although our Jupyter_HPC already offers generous compute capacity (see table below), extensive calculations should be translated into a sourcefile and started as a batch job from the regular scheduling system (SLURM) of the cluster.
The public Jupyter service description, terms of use and privacy statement can be found at the Jupyter service page of the University Computer Center Marburg.
A first overview on HPC usage can be found on our HPC service page. For details see MaRC3 documentation (requires login, accessible only from VPN or Marburg University network).
Access to Jupyter_HPC
Jupyter_HPC can by accessed at https://jupyter-hpc.uni-marburg.de.
The same prerequisites as for using the MaRC3 HPC cluster apply, including connection via VPN or university network. If you have no access yet, see FAQ on HPC access.
Starting and stopping a Jupyter server
After logging in with your (authorized) UMR staff account + password on the JupyterHub control panel, you can request a JupyterLab "server" on the MaRC3 cluster. If sufficient hardware is available, your requested Jupyter server will be scheduled and started within some seconds and you will be automatically redirected to it.
Depending on your needs, you can select one of five different job profiles. Your choice will determine the number of CPU cores, RAM and GPU resources that are reserved for you during the runtime of your session (see table below).
Table: Hardware profiles for Jupyter servers (March 2025)
| Profile | vCPU | RAM | Sub-GPUs | GPU-RAM | Max Users | Runtime |
|---|---|---|---|---|---|---|
| Standard | 4 | 4 GB | - | - | 12 h | |
| Extensive | 16 | 16 GB | - | - | 12 h | |
| Full | 32 | 64 GB | - | - | 12 h | |
| GPU10 | 16 | 16 GB | 1 | 10 GB | 4 | 12 h |
| GPU40 | 32 | 64 GB | 3 | 40 GB | 1 | 12 h |
In the spirit of fair use, you should only use the 16 or 32 core and GPU options when you actually need it. Otherwise, you block resources unnecessarily. Please note that only four concurrent sessions of the GPU10 profile and one session with the GPU40 profile are currently possible.
Please make sure to explicitly stop your Jupyter server once your task is complete.
In the JupyterLab menu go to: File > Hub Control Panel > Stop My Server.
If you just close the browser / tab or click on File > Lock Out this will leave your server running in the background and block others from accessing the shared hardware.
Figure: Terminating Jupyter sessions.
Using JupyterLab
After starting a JupyterLab session, you have access to the various file systems of MaRC3 and you are directed to your personal home directory. Using the launcher, you can start with a new Jupyter Notebook (*.ipynb), open a terminal session or create a new markdown or python file. An example is shown below.
Figure: Working on a Jupyter Notebook at MaRC3.
Jupyter training
You can find a recorded Jupyter introduction (plus presentation materials) held by Dr. Christian Berger as part of the HeFDI Data Talks 2024 here.
FAQ, Troubleshooting and Support
- FAQ can be found here or in the FAQ of this manual.
- Detailed information on Jupyter products can be found on the Jupyter website.
- Using the JupyterHub actually means using the HPC infrastructure. Potential issues might also be rather general HPC topics than specific issues with JupyterHub. It could be helpful to search also the service description and FAQ on HPC.