Proxmox on a Mini PC: LXC vs VM, Storage and Backups

Proxmox on a mini PC lets you separate Linux containers and full virtual machines, with storage and backups managed from one host. This guide starts after the host is installed and covers LXC versus VM allocation, storage layout, backups, and what the hypervisor costs in watts and RAM.

Proxmox is useful on one node when you need separate guest operating systems or guest-level management. If you are still choosing hardware, compare the mini PCs that actually run Proxmox well.

Install first

The install itself (BIOS settings, ext4 versus ZFS, the no-subscription repository and first updates) is covered step by step in the Proxmox VE 9.2 install guide for N100 mini PCs. The short version for a 16GB box: pick ext4 on a single NVMe so ZFS’s ARC doesn’t compete with your guests for RAM, and set a static IP at install.

The decision that defines a low-power Proxmox box: LXC vs VM

This is where mini-PC Proxmox lives or dies. Lean on LXC containers, not VMs.

• LXC containers share the host kernel. They boot in a second, idle at near-zero overhead, and a Debian LXC running a couple of services uses 60–150MB. On a 16GB box you can comfortably run 10–20 LXCs.
• VMs emulate full hardware and run their own kernel. Each idle Linux VM costs 0.5–1GB RAM and a measurable slice of CPU before it does any work. Four idle VMs can eat half an N100’s RAM.

The rule of thumb: LXC for everything Linux, VM only when you must — a non-Linux OS, a firewall appliance, a kernel module you don’t want touching the host, or strong security isolation for something internet-facing. A practical single-N100 layout:

Note the pattern: one LXC runs Docker and most services live as containers inside it. You get Proxmox’s snapshot/rollback on the whole Docker host without paying VM overhead per service. The team at DockerHomeLab has Compose stacks that drop straight into that docker-host LXC.

A common mistake is one VM per service. On low-power hardware that’s the fastest way to run out of RAM. Containers first, always.

The one place people reach for a VM unnecessarily is hardware video transcoding, on the assumption that reaching the integrated GPU requires PCIe passthrough. It does not: an LXC container gets the GPU as a device node, and the container route is both simpler and cheaper in RAM.

Match guest disks and backups to the mini PC’s actual storage interfaces. For dedicated appliance guests, consult the Home Assistant hardware guide or the pfSense mini PC guide for the USB and network requirements.

Storage and backups

Keep it simple on a single box:

• local-lvm / local: default, fine for guest disks on the NVMe. If it’s your only drive, the OS and guests share it.
• A separate SATA SSD as a directory storage for vzdump backups and ISO/template storage, if the box has a 2.5” bay. Keep the OS and guests on NVMe and add the SATA drive after install. Backing up to the same physical drive your guests live on protects you from nothing.
• Proxmox Backup Server is excellent but it’s another machine or VM. On a one-box setup, scheduled vzdump to the SATA drive plus an occasional copy to a NAS or external disk is the pragmatic minimum. If a real NAS is in your future, TrueNASGuide covers building the storage target these backups should eventually land on.

Set a nightly vzdump of every guest, keep 3–7 days, and actually test a restore once; the N100 install guide has the NFS target and restore-drill commands. An untested backup is a guess.

The watt cost of the hypervisor

People assume Proxmox burns a lot of idle power. It doesn’t burn much, but it isn’t free either. As a planning estimate for an N100 box (calculation inputs, not measurements by MiniLabHQ or a cited source):

• Bare-metal Debian + Docker, warm idle: ~8.6W
• Proxmox + one Docker LXC + DNS LXC, warm idle: ~9.4W

So plan on roughly 0.5–1W and a few hundred MB of RAM for the hypervisor layer on a single node — about $1–1.5/year in electricity. That’s a non-issue. The real cost of Proxmox on small hardware is RAM and complexity, not watts. For how to measure these deltas, see measuring mini PC idle wattage, and for the dollar framing, the power consumption cost math.

Tuning Proxmox for a low-power node

A few changes that help on small hardware:

• Disable the HA stack (pve-ha-lrm, pve-ha-crm) if you’re a single node. It does nothing useful solo and adds churn.
• Enable C-states in BIOS — Proxmox doesn’t disable them, but some mini-PC BIOSes ship with them limited. Same tuning as any other OS on the box, and part of the wider BIOS pass a 24/7 node wants alongside VT-d, Wake-on-LAN and power recovery.
• Don’t over-provision vCPUs. Giving a 4-core N100 guests that total 12 vCPUs just adds scheduling overhead. Match vCPU counts to what guests actually need.
• Budget the RAM before creating the guests. Memory, not cores, is what runs out first on a 16GB node — how many VMs an N100 can run gives a per-service allocation table and the symptoms of a host that is already over-committed.
• Turn off the subscription nag via the community script or repo swap so the UI stops interrupting you.

Bottom line

• Proxmox on one mini PC is worth it for snapshots, non-Linux guests, and learning — not as a default for a plain Docker stack.
• If you do run it, LXC for everything, VMs only when you truly must. That single discipline is what makes Proxmox viable on an N100.
• The hypervisor costs you RAM and complexity, ~1W of power — budget the RAM, ignore the watts.
• One box, one admin, all-Docker? Bare-metal Debian is simpler and lighter. Save Proxmox for when isolation or snapshots actually earn their overhead.

For the broader “what should I even be running” question that comes before the hypervisor decision, SelfhostRealm’s beginner guide sequences a sane first stack, and the $200 N100 build shows the bare-metal alternative side by side.

Related across the network

• The Beginner Homelab Guide: First 5 Services Worth Running — selfhostrealm.com