Servers

Currently, we have one GPU server

FHMLFYSLNX001 living in UNS50 F1.150a

Rules

• Make sure to monitor resource usage and be considerate
• htop can be used to monitor CPU and RAM usage
• nvidia-smi -l 1 shows GPU usage
• du -hs ./*/ shows the size of all folders in our /project drive. Make sure clean up regularly.

Connecting

• Simply connect via ssh using your normal UM credentials. The server is running on 3389.
• screen allows you to start a session that you can reconnect to, should the connection be lost, or you want to shutdown your computer.

Getting data on the server

Data for your projects should be stored in /projects. An administrator (i.e. Christian) will create a folder for you. Copy data to the server with your favorite scp client (e.g. WinSCP)

Using python on the server

You need to install your own python environments, for example with miniconda

VS Code remote

Install the Remote - SSH extension

Connect via SSH (bottom left corner) - Work on VSCode as normal

CUDA:

I want to use those cool GPUs, how do I do this? Run the following commands once:

echo "export PATH=/usr/local/cuda/bin${PATH:+:${PATH}}" >> ~/.bashrc
echo "export LD_LIBRARY_PATH=/usr/local/cuda-12.2/lib64\
                         ${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}" >> ~/.bashrc

Freesurfer

Run the following commands once:

echo "FREESURFER_HOME=/usr/local/freesurfer" >> ~/.bashrc
echo 'source $FREESURFER_HOME/SetUpFreeSurfer.sh' >> ~/.bashrc

The patient folder is set to /project/imaging/subjects

Graphical User Interface

Wait! All of this is in consoles. I want a User Interface.

• Using Windows Remote Desktop Connection, which requires a separate User and Password, you can log in and work remotely. Beware, this is a lot slower.