“Raspberry Pi boards are hard to come by, probably next year,” says Andreas Spiess, single board enthusiast and YouTuber, in his distinctive Swiss accent. He is not wrong. Spiess says he and his fellow Pi fans need a “strategy to survive”without new boards, so he suggests looking into one of the least exciting and most overlooked areas of computing: used enterprise thin client PCs.
Spiess Pi replacements, suggested and refined by many of his YouTube commenters and Patreon subscribers, are Fujitsu Futros, Lenovo ThinkCentres, and other small systems (some or all of which could be semantically considered “fat clients”or simply “mini PCs”). depending on your tastes and sense of retro grumbling). These are systems that you can easily find on eBay, refurbished on Amazon Renewed, or other sources of corporate and IT assets. They are generally in good shape considering their use and environment. And compared to single board systems for enthusiasts, many more are produced and replaced each year.
Of course, they have always been there, but now it makes sense to take another look at them. “Back to the future”as Spiess put it (by analogy, we’re not entirely sure it works).
Spiess’ journey to homemade Pi servers is quickly moving him away from modern NUCs, as they are too expensive and too powerful for a Home Assistant, PiHole, or even a multi-container system. He considers two types of thin clients that work. For a single-purpose Pi replacement, just about any Intel or AMD processor will do, and you’ll need 4GB of memory and an 8GB SSD. To run multiple Pi-scale projects out of the box, Spiess suggests finding a newer processor, 8GB of memory, and a 64GB or 128GB SSD (or upgrading either separately if possible). For his single-purpose projects, he found a €34 ($35) Futro with an AMD GX-222GC SoC; for a few containers, he got fourth-generation ThinkCentres i5 and i7 for 79 and 105 euros (82 and 109 dollars at the moment).
Then there is the software. Installing Home Assistant or similar Pi-centric images on a low-powered client requires a USB-to-mSATA adapter, or booting a Debian system from a working USB drive and then flashing the internal SSD with the image. For a machine with multiple VMs or multiple containers, Spiess uses Proxmox. He installs Home Assistant on one virtual machine, then IoT Stack on the AMD-64 version of Debian on another.
That a more powerful computer can mimic a less powerful computer is no surprise, but the difference in power consumption and processing may surprise some. Compared to the Raspberry Pi, the cheaper Fujitsu Spiess is actually slightly slower at compiling ESPhome than the Raspberry Pi 4, but the i5 and i7 were over three times faster.
Using a Spiess meter, I measured their power consumption. The two Raspberry Pis drew 12W mostly in idle mode, the Fujitsu 14W, and the i5 and i7 16-18W (with peaks up to 25W during high activity). Buying thin clients with newer processors and better power efficiency will lower those numbers even further, but Spiess’s calculations suggest savings of only about 24 euros ($25) per year, compared to the hundreds of dollars that cost more to buy them.
Clearly, more work is required to find the right balance of cost and size in the used thin client (or sometimes “mini PC”) market, and more work to launch one of the projects usually dedicated to the distinctive Pi platform. But supporting the used computer market is a net boon, especially if a machine perfect for repairs or home automation ends up in your network closet instead of your shredder. (Thanks to Hackaday for the video link).