There's a version of self-hosting that's about ideology — owning your data,
sticking it to Big Cloud. This isn't that post. This is the boring, arithmetic
version: I bought a small fanless x86 board and two hard drives, put
ZimaOS on it, and it now does the job I used to pay
several monthly subscriptions for. It sips power, it's been running for months,
and the thing I'd most want you to take away is the cost curve — because that's
the part the "self-host everything!" crowd usually hand-waves.
I'll show you the actual maths, what runs on it, the one genuinely unusual thing I
tried, and — because these writeups are worthless if they only show the wins — the
two ways it bit me and the one place I still pay for the cloud on purpose.
The hardware, and why "cheap and slow" is the point
The board is a low-power x86 mini-PC — the kind with an Intel Apollo Lake / N-series
chip, a couple of SATA ports, and no fan. Call it £100–160 depending on the model
and where you buy. Add two drives (mine were pulled from older machines, so
effectively £0, but budget £40–80 for a couple of second-hand ones). Total
out-of-pocket: somewhere under £150 for a lot of people, less if you have drives
lying around.
It is not fast. That is a feature, not a compromise. A home server spends 99% of
its life idle, waiting to serve a photo or a file. What you actually want is:
- Low idle power. Fanless N-series boards pull single-digit watts idling. At ~8W average and typical UK electricity, that's on the order of £15–20 a year to run. That number is the whole ballgame — hold onto it.
- Silence and no moving parts (beyond the drives), so it lives on a shelf and you forget it exists.
- Two drives, so you can mirror and survive one dying.
The OS is ZimaOS — a self-hosted, appliance-style OS built around Docker, with a
web UI, a built-in app store, and drive management. It's the successor to CasaOS
and aimed squarely at "I want a NAS-plus-apps box without becoming a Linux
sysadmin." You can absolutely run plain Debian instead; ZimaOS just removes a lot
of the fiddly first-day setup.
The cost maths, honestly
Here's the comparison that made me do it. The recurring cloud services a box like
this replaces, at rough UK consumer prices:
| What I was paying for | Typical monthly | Per year |
|---|---|---|
| 2 TB photo/cloud storage (the Google/Apple tier) | £7–8 | ~£90 |
| A file-sync / cloud-drive subscription | £8–10 | ~£110 |
| Odds and ends (extra storage, a backup app tier) | £3–5 | ~£50 |
| Total | ~£20/mo | ~£250/yr |
Against that, the self-hosted side:
| Cost | One-off | Per year |
|---|---|---|
| Board + two drives | ~£130 | — |
| Electricity (~8W) | — | ~£18 |
| Offsite backup (see below) | — | ~£10–25 |
| Effective | £130 up front | ~£30–40/yr running |
The shape is the point. The cloud line is flat forever — you pay ~£250 every
year until you die or cancel. The self-hosted line is a one-off hump then a
trickle: the hardware pays itself back in well under a year, and after that
you're running the same capacity for the price of the electricity plus a few quid
of backup. Year two onward, it's not close.
Two honest asterisks, because a table that hides them is lying:
- Your time isn't free. The setup is an afternoon, and there's occasional maintenance (more below). If you value a weekend at zero, self-hosting wins enormously; if an hour of your fiddling costs you £50 of billable work, do the sum for yourself.
- Capacity isn't the same as a managed service. You're buying the storage and the sync app — you're not buying someone else's 99.99% uptime, their geo-redundancy, or their support desk. Which is exactly why the backup line exists.
What actually runs on it
The box isn't a single app; it's a little stack, and each piece replaces a bill:
- Photo library + backup — a self-hosted photo app that auto-uploads from every phone in the house, does the "on this day" thing, and gives back the 2 TB I was renting. This is the single biggest subscription it kills.
- File storage / network shares — the NAS half. The two drives, mirrored, as the household's shared storage.
-
A reverse proxy — so the apps get clean hostnames and TLS instead of
ip:porteverywhere. (I run a community proxy image rather than the default one, but any will do.) - A dashboard — one page listing everything with health dots, so the family has a front door and I can see at a glance what's up.
- Edge security agent — a lightweight intrusion-detection agent feeding a central policy, so the box isn't naked even though most of it never faces the internet.
None of it is exotic. That's deliberate — the boring stack is the one that's still
running in a year.
The one genuinely unusual thing I tried
Because a fanless board is too boring, I tried to give it a GPU. These little
boards expose a PCIe lane, and with a riser you can physically attach a
low-profile card — I hung an old workstation GPU off it to accelerate photo
thumbnailing and the odd transcode.
It half-worked, and the ways it didn't are the useful part:
- The power input is USB-C, not a barrel jack — which matters enormously the moment you add a second power source for the GPU, because now you've got two supplies and a beautiful opportunity for a ground fault between them. I found that opportunity.
- Old GPUs age out of software. The photo app's newer version dropped support for the card's GPU generation, so the acceleration I'd wired up quietly stopped being an option after an update. The card became a space heater.
I keep it documented because it's a great example of the homelab tax: the
interesting mod is 20% soldering-iron fun and 80% discovering why the tidy
diagram doesn't survive contact with a real power rail. For most people, the
honest advice is: skip the GPU, let the CPU do the thumbnails slowly, it's
fine.
How I keep it alive (and where I still pay the cloud)
A cheap board and two drives only "keeps everything alive" if you're honest about
the ways it can die. Mine:
Mirror the drives. Two drives configured so one can fail without data loss.
This is table stakes and it's the reason for two drives, not one.
The 3-2-1 rule still applies, and this is where cloud earns its keep. A mirror
protects against a drive dying. It does nothing against the box being stolen,
the flat flooding, or me rm-ing the wrong thing. So the irreplaceable stuff —
the photos, mainly — is backed up off-site to cheap cloud object storage,
encrypted, for a few pounds a month. This is the nuance the "ditch the cloud"
crowd skips: I didn't eliminate the cloud, I demoted it. It went from being my
primary storage (expensive, per-app, £250/yr) to being my backup of last
resort (cheap, bulk, ~£10–25/yr). That's the actual win — not "no cloud," but
"cloud in the one role it's genuinely best at and cheapest for."
Watch it, and alert on absence. A dashboard tile going red is nice, but the
failure that hurts is the backup that silently stops running. So the backup job
reports when it succeeds and something screams if an expected run doesn't happen
at all. A quiet week is not proof of health; I make the box prove it's alive.
Beware the appliance-OS "helpfulness." The one recurring bite: the friendly
app-manager layer that makes ZimaOS/CasaOS easy will, on an update or a reboot,
sometimes decide to "tidy up" a container it didn't create — twice it removed a
dashboard I'd deployed by hand. The fix was to deploy the things I care about as
plain, declared Docker containers I control, not through the pretty UI, so the
appliance layer keeps its hands off them. Convenience layers are great until
they garbage-collect something you needed.
Should you do it?
Run the decision, not the ideology:
- Storage + photos are your big cloud bill → this pays for itself fast. A sub-£150 box kills a ~£250/yr habit and then costs pennies to run. Clear win.
- You'll enjoy (or at least tolerate) an afternoon of setup and rare maintenance → go.
- You want zero maintenance and someone to call when it breaks → keep paying the cloud; that's literally what you're paying them for, and it's a fair deal for some people.
What the cheap-board approach really buys you isn't just money — it's that the
cost stops recurring. You trade a bill that renews forever for a one-off lump and
a trickle of electricity. Do that once for the thing that's costing you the most
per month, keep the cloud only as the cheap off-site safety net, and the maths
stops being close.
The unglamorous truth: a silent little board on a shelf, two mirrored drives, and
a boring stack of containers will quietly do what several subscriptions were doing
— for the price of a coffee a month in electricity, and one honest cloud backup
bill you're actually happy to pay.
Homelab notes from someone whose "server" costs less to run per year than a
single month of the subscriptions it replaced. Prices are illustrative UK ballpark
figures — run your own numbers, the shape of the curve is what matters.
🤖 Drafted with AI assistance from my own homelab notes, logs and repos, then reviewed and edited before publishing.
Top comments (0)