Jupyter server for DTU HPC/3D-imaging-center

_README.md

Jupyter server for DTU HPC/3D-imaging-center

This gist is a small guide for getting a Jupyter server running on DTU HPC on a GPU node.

client.sh

#!/bin/bash
# Client, can be run on any node

# read config
cd ~/jupyter
. config.conf

# connect to the jupyter server
ssh $USER@login2.hpc.dtu.dk -NL$PORT:$SERVER_HOSTNAME:$PORT

config.conf

PORT=44012
CONFIG=config.conf
USER=s204163
GROUP=group_Curlew
SERVER_HOSTNAME=na

server.sh

#!/bin/bash
# Server, remember to switch to appropriate node before running; linuxsh, a100sh, sxm2sh, etc.. 
# see: https://www.hpc.dtu.dk/?page_id=2129

# read config
cd ~/jupyter
. config.conf

# set the hostname in the config, clients need to connect to this
sed -i "s/^SERVER_HOSTNAME=.*/SERVER_HOSTNAME=$HOSTNAME/" $CONFIG

# activate env, this can be any enviornment, here it is a conda env
cd /dtu/3d-imaging-center/courses/
.  conda/conda_init.sh
conda activate env-02510

# expose "all" GPUs. Please note that devices 2,3 will only be available on sxm2sh since it has 4 GPUs. Use at own risk
export CUDA_VISIBLE_DEVICES="0,1,2,3,4"

# start notebook server at the location with your files
cd /dtu/3d-imaging-center/courses/02510/groups/$GROUP/$USER/
jupyter notebook --port=$PORT --ip=$HOSTNAME --no-browser