Remote GPU Development Environment

Transform your gaming PC or any Linux machine into a remotely accessible GPU development server. Connect securely from anywhere using Cloudflare tunnels — no port forwarding, no static IP, no router configuration required.

Designed for deep learning training, data science, and GPU-accelerated workloads. Accessible from Solveit, VSCode, or any SSH client.

---

What This Does

• Converts a Windows gaming PC or Linux machine into a remote GPU server
• Exposes it securely via Cloudflare tunnels
• Manages multiple persistent Jupyter kernels — one per user or client
• Keeps kernels alive across sessions so you can resume work anytime
• Automatically cleans up inactive kernels at 10pm daily
• Works with Solveit, VSCode Remote SSH, and Mac/Linux terminals

---

Installation

Windows Setup (WSL)

Prerequisites

• Windows 10/11 with Administrator access
• An SSH public key for remote access
• A Cloudflare account with a registered domain

Run the Setup Script

Run as Administrator in PowerShell:

 irm https://raw.githubusercontent.com/rleyvasal/gpu-dev-setup/main/setup-windows.ps1 | iex

On first run, you will be prompted for:

• WSL distro name (auto-detected if only one exists)
• Linux SSH port
• SSH public key
• Cloudflare domain and tunnel name
• Project name

These values are saved to ~/.config/gpu-dev/client-config.json so subsequent runs skip the prompts.

What the Script Automates

Step	Description
WSL 2	Installs WSL and your chosen Linux distro
OpenSSH Server	Host key generation, pubkey auth, password auth disabled
Windows Hello	Disables passwordless requirement for SSH compatibility
SSH key	Adds your public key to administrators authorized keys
Firewall	Opens ports for SSH (22 and custom Linux SSH port)
Power settings	Disables auto-sleep and hibernate (AC and DC)
WSL idle timeouts	Creates .wslconfig to prevent WSL from shutting down
WSL startup	Scheduled task to start WSL at boot/login
Config files	Written for both Linux and Windows sides
Linux setup	Triggers setup-linux.sh inside WSL automatically

Manual Steps Required

1. First WSL launch — If your distro is freshly installed, launch it once and create your Linux user before running the script.

2. Cloudflare tunnel login — After the script runs, if no tunnel is found, you will be prompted to run cloudflared tunnel login inside WSL. This requires browser-based authentication with your Cloudflare account.

3. Adding SSH keys from new clients — The script only adds the key provided during setup. To add keys from additional clients (e.g. solveit), manually append them on Windows:

   Add-Content "C:\ProgramData\ssh\administrators_authorized_keys" "<public-key>"
   icacls "C:\ProgramData\ssh\administrators_authorized_keys" /inheritance:r /grant "SYSTEM:F" /grant "Administrators:F"
   Restart-Service sshd

---

Native Linux (Standalone or WSL)

The Linux setup script supports two modes:

• Standalone — install directly on a Ubuntu/Debian machine with a GPU
• WSL — called automatically by setup-windows.ps1

When running in WSL mode, the script configures both Linux and Windows SSH access through a single Cloudflare tunnel. When running standalone, it configures only Linux SSH access.

Before You Start (Standalone Only)

If running standalone (not via the Windows script), complete these steps first:

1. Install cloudflared if not already installed:

   wget -q https://github.com/cloudflare/cloudflared/releases/latest/download/cloudflared-linux-amd64.deb -O /tmp/cloudflared.deb
   sudo dpkg -i /tmp/cloudflared.deb

2. Authenticate with Cloudflare:

   cloudflared tunnel login

3. Have your SSH public key ready. This is the public key from the client machine (e.g. solveit, Mac) that will connect to this server.

   • For Solveit: cat /app/data/.ssh/id_*.pub
   • For Mac/Linux: cat ~/.ssh/id_ed25519.pub

Run the Setup Script

curl -fsSL https://raw.githubusercontent.com/rleyvasal/gpu-dev-setup/main/setup-linux.sh -o /tmp/setup-linux.sh && bash /tmp/setup-linux.sh

The script will prompt you for:

• Your SSH public key
• Your Cloudflare domain (e.g. mydomain.com)
• Your tunnel name (e.g. gpu-dev)
• Your project name (e.g. myproject)

Note: When called from the Windows script, these values are passed automatically via config and the script runs in non-interactive mode.

What the Script Automates

Step	Description
SSH server	Configured on port 2222 with public key auth only (password auth disabled)
Firewall	UFW rules for SSH ports (standalone only; skipped in WSL)
Sleep/suspend	Disabled for always-on GPU server (standalone only; handled by Windows host in WSL)
Python venv	Created with uv, includes ML packages (torch, torchvision, torchaudio, numpy, pandas, scipy, scikit-learn, matplotlib, plotly, pillow, tqdm, httpx, requests)
Cloudflare tunnel	Created, DNS routed, and runs as a systemd service
Cloudflared symlink	Created at ~/.local/bin/cloudflared for client compatibility
Kernel manager	Installed with daily cleanup timer (10pm)
Client config	Written to ~/.config/gpu-dev/client-config.json

WSL Mode Differences

When running in WSL, the script additionally:

• Creates a DNS route and ingress rule for Windows SSH access (CF_HOSTNAME_WIN)
• Skips UFW firewall configuration (handled by Windows)
• Skips sleep/suspend configuration (handled by Windows host)

Notes

The default venv includes common ML/data science packages. Sleepnow alias | Adds sleepnow command to PowerShell profile for manual sleep |