VS Code Remote Setup on RTX 5060 Ti 16GB

VS Code Remote-SSH is the developer-ergonomics sweet spot for GPU work – your editor runs locally on your laptop but everything (interpreter, language server, debugger, terminal) runs on the remote box. This guide gets VS Code talking to a RTX 5060 Ti 16GB server on our dedicated GPU hosting, with Python, CUDA, and port forwarding for vLLM.

Contents

• Why Remote-SSH
• SSH config
• First connect
• Extensions and interpreter
• Port forwarding
• Performance tips

Why Remote-SSH

SSH Config

On the laptop, edit ~/.ssh/config:

Host gpu-5060ti
    HostName your-server.gigagpu.com
    User ubuntu
    IdentityFile ~/.ssh/id_ed25519
    ServerAliveInterval 60
    ServerAliveCountMax 3
    Compression yes
    # Tunnels opened automatically on connect
    LocalForward 8000 127.0.0.1:8000   # vLLM OpenAI
    LocalForward 8888 127.0.0.1:8888   # Jupyter
    LocalForward 6006 127.0.0.1:6006   # TensorBoard
    LocalForward 9100 127.0.0.1:9100   # Prometheus node-exporter

Test from a terminal: ssh gpu-5060ti nvidia-smi. You should see the RTX 5060 Ti with 16 GB VRAM.

First Connect

1. Install VS Code and the Remote – SSH extension (ms-vscode-remote.remote-ssh)
2. Command Palette (Ctrl/Cmd+Shift+P) -> Remote-SSH: Connect to Host -> gpu-5060ti
3. VS Code installs its 80 MB remote server automatically on first connect (~30 s)
4. Open the remote folder you want to work in

The status bar shows SSH: gpu-5060ti when connected. All terminals, debugger and file edits now happen server-side.

Extensions and Interpreter

Extensions install either locally or remotely. Install on the SSH host:

Create a venv on the server and point VS Code at it:

# On remote (via VS Code terminal)
python3.11 -m venv ~/venv
source ~/venv/bin/activate
pip install torch --index-url https://download.pytorch.org/whl/cu126
pip install transformers accelerate vllm ruff

# Then Ctrl/Cmd+Shift+P -> Python: Select Interpreter -> ~/venv/bin/python

Port Forwarding

Two ways. Either via the SSH config LocalForward lines above, or via VS Code’s Ports panel – auto-detected when a local server binds a port. Useful ones:

Test vLLM forwarding end-to-end:

# Laptop terminal
curl http://localhost:8000/v1/models | jq
curl http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"llama","messages":[{"role":"user","content":"hello"}]}'

Performance Tips

• Enable SSH Compression in the config when on slow links
• Exclude node_modules, .venv, *.bin from workspace file watching – huge perf win
• Use ControlMaster auto and ControlPath for instant reconnects
• Keep the remote server on 1000 MB/s ethernet – Remote-SSH traffic is small but bursty

See also: Jupyter setup, Docker CUDA setup, Ubuntu driver install, vLLM setup, first-day checklist.