Your desktop can be more than a PC

A gaming rig sits idle most of the day. Mine now runs a media server, a NAS, and a handful of services I actually use — without dealing with the Windows headache.

When you build a PC, the goal is to get the most out of your budget while keeping a sensible upgrade path. The problem is that one machine is rarely enough anymore. As someone who straddles research computing and everyday work, you end up needing Linux for experiments, Windows for compatibility, and something to share files cleanly across both. The usual answers — put up with a slow laptop, buy several devices, or pay for cloud instances you barely use — all have obvious costs, either in money or in patience.

The better answer turns out to be the one already sitting under your desk. A modern CPU runs at two or three percent utilization most of the day. The electricity is already on. If you have an older machine collecting dust, it might quietly replace four EC2 instances without breaking a sweat.

The setup

I timed the build well — RAM and SSD prices happened to be near historic lows. The CPU is a Ryzen 7 7700, which turned out to be more than sufficient. Instead of one powerful GPU I went with two modest ones: a 3050 drives the Windows desktop across two monitors, while a 5060 Ti handles GPU compute on a headless Linux VM. Consumer cards don't support hardware virtualization anyway, and Windows quietly claims a slice of your GPU even when you're trying to benchmark something. Separating the two workloads solved both problems at once.

The hypervisor layer is Proxmox VE, an open-source platform built on KVM and LXC. It exposes a clean web UI and a solid API, so you can spin up, suspend, or snapshot any VM from a browser — or from your phone, from anywhere. One practical caveat: if your server connects to your home network over Wi-Fi rather than ethernet, you need to install wpasupplicant manually before Proxmox can reach the internet. The trade-off is that you also lose the default Linux bridge, which means your VMs end up in their own subnet rather than on your home LAN. In practice this is fine — forward the ports you actually need, and everything else stays isolated behind one extra hop.

What actually runs on it

The most immediately useful service is a NAS. HDDs are not obsolete — a couple of them plus TrueNAS gives you a network share that every device on the house can mount natively, regardless of whether it runs Windows, macOS, or Linux. Each ecosystem has its own file-sync story, but none of them talk to each other willingly. A shared network folder sidesteps the whole problem.

On top of the NAS sits Jellyfin. Rather than casting a laptop screen to the TV or copying files to a USB stick, any media on the NAS streams directly to whatever is on the TV. The iGPU inside the Ryzen 7700 handles hardware transcoding without issue, so a discrete GPU is not required for this at all.

The headless Linux VM is where compute jobs actually run. The 5060 Ti is passed through directly to it, so GPU utilization is native — no virtualization overhead, no sharing with the display stack. Tailscale runs on the VM and takes about five minutes to set up; after that, the machine is reachable from anywhere on any device. From a laptop on a coffee-shop Wi-Fi or an SSH session on the phone, it behaves exactly like a local connection. This is, effectively, a private cloud instance that costs nothing to run once the hardware exists.

The Windows VM is last and, by resource consumption, first. For emails, papers, and anything that genuinely needs a polished GUI, a full desktop environment is hard to beat. I have tried Linux desktop environments and tiling window managers; the display driver support for my monitor configuration was never quite right, and the friction accumulated. Windows as a VM, isolated from everything else, turns out to be the honest solution.

What surprised me

Mostly, how quickly it disappears. Once everything is running, there is nothing to do. The VMs start on boot, the services are reachable, and a week goes by without touching the Proxmox UI. The setup becomes infrastructure rather than a project, which is the goal. The one genuine surprise is how little the system cares about anything except a power outage — that is the only thing that actually interrupts it. The other thing worth flagging is upgrades: Proxmox major version bumps require careful reading and a bit of patience. Day-to-day, nothing; but treat upgrades with the same respect you would a production server.

Was it worth it

Yes, with an honest caveat: the initial weekend of setup is real work, and you need to enjoy that kind of work for the answer to be yes. If debugging a Wi-Fi bridge issue at midnight sounds like a bad time, this is not for you. If it sounds like a Saturday well spent, the return is good. After the setup cost, the ongoing expense is essentially zero — the hardware was already there, the electricity was already running, and you now have a NAS, a media server, a GPU compute node, and a Windows desktop running simultaneously on one machine that used to do one thing.