Non-RCI Servers

These are smaller, flexible servers that you access via SSH. They’re ideal for:

• Getting started with GPUs
• Running small-scale or interactive jobs
• Debugging with PyCharm or Jupyter

🔐 VPN & SSH Access

• Use OpenVPN (recommended over Tunnelblick on macOS).
• Connect with:

  ssh username@server_name.felk.cvut.cz
  

  Example servers: duda, ptak. Some machines (like cmp-grid79) act as jump hosts.

🔑 SSH Keys

• Generate and manage SSH keys
• Copy SSH key to remote server

📁 Data & Storage

• Main locations: /home, /datagrid/personal/username, and shared /datagrid/public_datasets
• Avoid duplicating large datasets; use symbolic links if needed.
• Storage docs

⚙️ Python and Other Software Setup

See Python environment setup.

🚀 Running Experiments

• Check GPU usage with nvidia-smi
• Set GPU visibility:

  export CUDA_VISIBLE_DEVICES=0  # Or 0,1 or 4,7
  

• Use screen or tmux to keep processes alive after logout:

  screen -S myjob
  Ctrl+A D  # Detach
  screen -r myjob  # Reattach
  

🐞 Debugging & Remote Tools

• Remote SSH interpreter (PyCharm)
• Multi-hop SSH (if server is behind a jump host)
• Remote Jupyter setup

🔄 File Sync

• Use PyCharm deployment or:

  rsync -avz myfile.py username@server:/home/username/project/
  scp model.ckpt username@server:/datagrid/personal/username/
  

• SSHFS works but can be tricky on macOS.