The setup I’ve been running is five Windows machines spread across two locations. They’re powerful, they’re fast, and every single one of them is off at some point. That’s a problem when you want to run anything reliably.

So I added an Orange Pi 5 between them.

What it is

Orange Pi 5, RK3588S SoC, 8GB RAM, Ubuntu 22.04 ARM64. Small, fanless, draws almost nothing. It lives on the network and never gets switched off.

That’s the entire point. Everything else in the setup is optional. The Pi is always there.

What it does

Wake-on-LAN for the fleet. All the heavy machines – the workstations, the edit rigs, the AI agent box – can be woken remotely. The Pi sends the magic packets. One SSH command from anywhere: ssh opi5 wake schnitt-rechts and the RTX 4080 machine spins up. No need to be home.

Fleet watchdog. A systemd timer runs every 10 minutes, pings all hosts, writes status to a log. When I log into the Pi, the MOTD shows me what’s alive and what isn’t. A Telegram bot (@SaschasSocialMediaBot) sends alerts when machines go down or come back up, and a morning summary at 08:00.

Ollama fallback. The Pi runs llama3.2:3b and nomic-embed-text locally. Slow – around 8 tokens per second – but it’s there when every other machine is off. Enough for lightweight queries and embeddings without touching the cloud.

Pi-hole. Network-wide DNS filtering. The FritzBox upstream DNS points to the Pi. Ads and trackers blocked for every device on the network, with automatic fallback to public DNS if the Pi ever goes down – so the household stays online either way.

SSH from anywhere. A Cloudflare tunnel handles remote access without a VPN. ssh opi5-cf connects to opi5.manke.app from anywhere in the world. Key-only, no password login.

Backup relay. Nightly at 03:30, configs and critical data get pushed to the NAS. The Pi coordinates that without needing any of the workstations to be awake.

What changed in the architecture

Before: five powerful machines with a collective uptime problem. After: five powerful machines plus one small one that’s always on and handles everything that needs to run continuously.

The heavy compute still happens on the x86 boxes. The Pi just makes sure I can reach them, wake them, monitor them, and have a minimal fallback when they’re all off.

It cost around 80 euros and took an afternoon to set up.