What the home lab looked like
The starting point was a pile of used enterprise gear. The author says several 1U servers ran Proxmox. A prebuilt NAS running TrueNAS held six 4 TB drives. By his estimate the lab alone drew just under 300 W, and the household idle load sat around 600 W. He puts that at nearly $5 a day, or roughly $150 a month.
These are the author's own numbers. He gives no metering method, measurement period or utility rate.
A quick sanity check on the cost: 300 W running around the clock is 7.2 kWh a day. At $5 a day, that implies a rate near 69 cents per kWh. That's very high by most standards, so the cost figure probably covers more than the lab's draw, or reflects an expensive tariff. Treat it as a rough household estimate, not a lab-only cost.
Phase one: everything on one Threadripper
The author took an idle AMD Threadripper and put a fresh Proxmox install on it. His migration, as he describes it:
- Back up every VM and LXC on every old node, using Proxmox's built-in backup rather than Proxmox Backup Server.
- Copy the backup files to the new machine with secure copy.
- Restore and start the guests on the new host.
- Run the NAS as a TrueNAS VM.
- Replace the six 4 TB drives with two 14 TB drives in RAID1. The old six-drive array offered 14 TB of usable space, so nothing was lost.
Household members noticed only an hour or two of downtime, he says.
His reported result was 120–150 W at the wall for the whole lab. Against a baseline just under 300 W, that is a cut of at least 50%, and up to about 60% at the low end. The household idle load fell from 600 W to 450 W, a 25% drop. He doesn't explain why the household figure fell by less than the lab's, though other loads staying put would account for part of it.
He also added a Radeon RX 7900 XT to run local models with Ollama and Open WebUI. He says the card gave him 20 GB of VRAM. He doesn't name the models or give any performance numbers.
What Proxmox documents about the backup route
Proxmox's feature documentation backs up the general approach. The platform's integrated backup tool, vzdump, can be started from the GUI or the command line. Proxmox says these backups are always full backups, with the guest configuration and all data. The tool also takes consistent snapshots of running containers and KVM guests.
Proxmox Backup Server is optional. It adds incremental transfers and client-side encryption, which a one-time migration doesn't need.
A backup archive is not a live transfer. The author doesn't document exact commands or per-guest validation, so don't read his account as a guaranteed recipe. Verify each backup before you wipe an old host. Test each restored service before you declare victory.
The big caveat: TrueNAS as a VM
This is the part most worth your attention. The author says only that "a TrueNAS VM" served as the NAS. He doesn't say whether the disks reached it by controller passthrough, by single-disk passthrough or as plain virtual disks.
That detail decides how safe your data is. The Proxmox community and other XDA reporting point the same way:
- A Proxmox forum answer says you should pass through a controller, never individual disks.
- With disk passthrough instead of PCI passthrough of the controller, TrueNAS only sees virtual disks, so it can't monitor drive health or spin disks down.
- XDA's own TrueNAS-on-Proxmox write-up says iXsystems recommends passing the entire SATA controller, which gives the VM the same control over the drives as bare metal.
- That article also warns that an extra virtualization layer that reorders write operations could make ZFS pools unusable.
- Controller passthrough needs IOMMU support: AMD-Vi or Intel VT-d.
The TrueNAS hardware guide, as relayed in the research notes, says virtual disks for a pool are not supported for regular or critical data. TrueNAS's installation documentation lists minimum VM settings of 8 GB of RAM, a 20 GB OS disk, at least one data disk and bridged networking. Those are install minimums, not a performance or protection guarantee.
One more side effect: with whole-controller passthrough, the drives vanish from the Proxmox host. You also lose host-level snapshots of the data. A Proxmox backup of that VM would capture only its boot disk.
The takeaway is that this migration working for one household doesn't prove every TrueNAS-in-Proxmox setup is safe. Before copying this design, find out how your disks will reach the VM.
Phase two: splitting it back up
The author's last arrangement isn't a single box. His main complaint with the one-server design was the shared failure domain: reboot the host and the NAS goes down with it.
His final setup:
| Role | Hardware | Reported draw |
|---|---|---|
| Compute (most VMs and LXCs) | Mini PC | "almost nothing" |
| Storage | TerraMaster T6-423 NAS with the two 14 TB drives | About 20 W |
| Camera recording and object detection (Frigate) | Fujitsu Siemens small-form-factor PC with a Core i5-7400 | Low, no figure given |
He says all three together draw under 100 W. That would beat the Threadripper phase's 120–150 W. The figure is again unmetered and undated, but it fits the pattern: a modern mini PC and a purpose-built NAS idle far below a Threadripper platform. He keeps the Threadripper in reserve.
Note the hardware's rated power as well. AMD rates the RX 7900 XT at 315 W of typical board power, according to the research notes. That's a rating, not what the card pulled in his lab. It does show why a single-box number depends heavily on GPU activity and configuration.
What to take from it
- Fewer boxes isn't automatically better. Consolidation cut the lab's draw by at least half. It also tied storage availability to the compute host.
- Idle efficiency matters more than raw capability. The biggest savings came from moving off old rack servers. The last step, to a mini PC plus NAS, saved more again.
- Separate by blast radius. Storage, compute and recording each got hardware suited to its job and its failure behavior.
- Plan storage first. If you virtualize a NAS, pass the controller through, back up the NAS config and keep a copy of the data off that host.
- Measure yourself. A cheap wall-socket meter beats anyone's estimate, including these.
Overall this is a useful personal case study, not a benchmark. The measurements are self-reported, and the storage setup that matters most is the one detail left out. For anyone weighing a rack of used servers against a few small, efficient machines, the lesson is to treat power draw and failure domains as design inputs from the start.