Building a home lab has been a lot of fun. Moving from a pre-built network attached storage (NAS) device to an enterprise setup and server rack, I learned a lot about how to best configure a smart home and what I need in a home office. It turns out that less can be more, so I replaced those enterprise beasts with singles AMD Threadripper server. This thing absolutely flew with 64 cores, but the 150W idle power draw was just ridiculous. At that time, I looked into mini computers, but instead of settling for these small devices, I ended up with several compact computers.
They are nothing special. 6th and 7th gen Intel processors, custom motherboards, and other parts, but I can freely max out the system memory, and with a 3.5-inch drive and the option to install some expansion cards, the I/O isn’t terrible. One of these Proxmox nodes has an Intel Core i5-7400 CPU, 32 GB of DDR4 RAM and a small 256 GB SSD. It’s powerful enough to run Nextcloud, Jellyfin, Home Assistant, and Frigate, but draws about 25 watts from the wall. This is the total power of the system. Not only do I save a lot on battery drain, but it’s also quieter and takes up less space.
Choosing the right compact computers
Just choose one of everything you need
There are countless compact systems available on classifieds sites. I chose the Dell Vostro 3268 and the Fujitsu Esprimo D556/2/E85+, two very similar systems with different plastic shells. There were two reasons for this: they are reliable and energy-efficient systems with the same quad-core processor, the Intel Core i5-7400. Fortunately, due to the RAM situation, I already had to source modules with 16GB DDR4 for the Dell and 32GB for the Fujitsu. These two nodes running Proxmox make up quite a home lab cluster and allow me to partition containers and VMs alike.
Getting Proxmox up and running is incredibly easy. Once the two nodes were live, I logged into the web management interface for the first node and created a cluster with the second. I could then use both GUIs to manage the entire data center as well as smartphone applications. All this makes everything quite simple to maintain. I determined what was needed to work and divided everything between the existing nodes. A compact computer, Fujitsu, will be tasked with working Medusahome page, House helper, The next cloudand Frigatetaking advantage of the available SATA port for a large control hard drive.
The Intel Core i5-7400 has integrated graphics with transcoding support to handle media streaming packages like Plex and Jellyfin. This almost makes the chip good for controlling home surveillance with the Frigate and some connected IP cameras. I’ve got an Nvidia T1000 on hand to handle some of that workload, in case things get complicated with security extensions, but even with four 2K cameras running detection, it’s pretty solid, reliable, and responsive. Nextcloud and Homepage are quite lightweight and don’t consume a lot of resources.
Overall, CPU usage is in the low 10s and mid 20s, mostly due to Frigate, although it looks like one node is running some heavy stuff.
You don’t need a lot of power
Many server builds are overkill
There’s a reason why minicomputers are recommended by the home lab community. They’re low-power, efficient, and pack enough punch for some of the heavier workloads. Look, don’t get me wrong, use it Threadripper 9980X was fun There’s something about having essentially unused processing power on hand for just about anything. The headspace was ridiculous. I tried to cram as much as possible into this system with Proxmox as the foundation and still barely made it in general use.
I have an older Minisforum U850 with Windows 10 and Intel Core i5-10210U, 16GB DDR4 RAM and UHD 630 graphics. It packs a punch and will give even my larger compact desktops a run for their money. The U850 measures 127 × 127 × 53.1 mm and can be mounted behind the monitor to keep it out of the way. The same goes for home lab use. Do you have a small data cabinet and dedicated DIY router for networking gear? A mini PC can sit there. Trembling enough, I managed to get both compact computers into the cabinet.
It frees up a lot of space, gets everything out of the way, and if you’re lucky enough to have a garage or outbuilding with power and mains access, it doesn’t make any noise for the home. If you’re just starting out and want to get your smart home and home lab up and running, it’s tempting to go big and loud by picking up high-performance hardware at a fraction of the MSRP, but that can lead to some issues. Parts are rarely available as new, and you may even be required to purchase new systems to remove specific components. And it’s all too much. I have a 10Gb trunk that connects my office to the house and garage.
Affordable switches manage connections and OPNsense runs on a low-cost, passively cooled firewall PC. It’s all just crazy, but that’s the point. The entire network is reliable, robust, and usually consists of parts that can be replaced without much trouble.
Self-hosting doesn’t require much
Home Assistant, Plex, Jellyfin, Frigate, Nextcloud and other self-hosted packages usually do not require a lot of system resources. It’s only when you move on to running your native large language models (LLM) and other intensive tasks that more powerful hardware is required, but even then, it’s possible to get by with low-powered devices like mini and compact PCs. If I need more capacity to run something else, either a new CPU, some more RAM, etc. I may look at upgrading the one I have with or get another node for the cluster.
