Skip to content

Instantly share code, notes, and snippets.

@beratkurar
Last active January 12, 2025 12:06
Show Gist options
  • Save beratkurar/d898ca97466312293ff70037df3b5c06 to your computer and use it in GitHub Desktop.
Save beratkurar/d898ca97466312293ff70037df3b5c06 to your computer and use it in GitHub Desktop.
This gist shows how to use Tel Aviv University GPU servers
  1. General information about connecting Tel Aviv University servers is here.

  2. How to use SLURM is here.

  3. Linux servers are:

nova.cs.tau.ac.il
soul.cs.tau.ac.il
op-controller.cs.tau.ac.il
c-[001-008].cs.tau.ac.il
  1. Windows server is:
terminal-1.cs.tau.ac.il
  1. The Linux servers by default starts with tcsh. Switch to bash using the following command:
bash
  1. Useful software are:
gedit
matlab
python
pycharm
  1. op-controller.cs.tau.ac.il contains the SLURM. Sending SLURM jobs requires special permission. Useful SLURM commands are:
sinfo
squeue
srun
sbatch
  1. c-[001-008].cs.tau.ac.il contain directly accessible GPUs. Connect to a client cluster:
ssh c-008.cs.tau.ac.il
  1. View the available GPUs:
nvidia-smi
  1. View the users who are occupying a GPU:
ps -u -p $pid
  1. Create private conda environment at c-008.cs.tau.ac.il using the instructions here. All the installations must be done on the NetApp storage for example '/home/nachum/berat'

  2. Connect to c-008.cs.tau.ac.il and start bash

ssh c-008.cs.tau.ac.il
bash
  1. See the available conda virtual environments.
conda env list
  1. Create a virtual environment.
conda create -n venvtf tensorflow-gpu
  1. Deactivate the virtual environment.
conda deactivate
  1. See the cuda version installed on the GPU node and see the GPUs available and the nvidia-driver version.
nvcc --version
nvidia-smi
  1. Activate the virtual environment
conda activate venvtf
  1. See that tensorflow 2 is installed
conda list
  1. Install the other packages that you need
conda install opencv
conda install jupyterlab
conda install -c anaconda ipykernel
  1. Start the jupyter server in background
jupyter lab --no-browser --port=8080 &
  1. Open a local browser and use the following address to open the jupyter notebook
http://c-008.cs.tau.ac.il:8080
  1. Or open an ssh thunnel to tau server and open the jupyter client at the local host. Hence the Jupyter traffic runs through ssh.
ssh -N -L 8080:localhost:8080 berat@c-008.cs.tau.ac.il:8080
localhost:8080
  1. List all running programs
ps -e
ps
ps -u berat
  1. Kill a process by PID
kill PID
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment