There are many advantages of the 1l mini PCs for home server applications: Obviously they are very compact, but more interestingly: The idle power consumption can be almost as low as current Raspberry Pi 5s acchieve, i. e. 5-6W*, all while avoiding many of their limitations**. In many cases the low power consumption is associated with quiet operation, which is especially important if you don’t have access to a dedicated server room in the basement or at least an acoustically isolated closet. Also, used mini PCs are available in large quantities at reasonable prices. This is why there is a ton of information and data out there, covering hardware expandability, compatibility with hardware and software and other aspects of converting mini PCs into capable home servers.
After my generally positive experience with a M720q as a lab Desktop PC (minus a worn fan) I chose used M920q as my proxmox cluster nodes. One of the main advantages of the ThinkCentres vs. many (not all!) other mini PC models is the PCIe x8 slot that can be used with SFF expansion cards like GPUs or NICs. While the M720q is built around the more limited B360 chipset, the M920q uses the Q370 chipset and brings features like Intel vPro. Both models have one disadvantage in common: Despite having footprints for two M.2 2280 slots for NVMe SSDs, only one is populated – the second slot is reserved for the much rarer and significantly more expensive M920x variant.
But fear not, soldering/rework equipment and skills presumed, one can convert the M920q model into something very similar to an M920x. (As far as I could see there is no UEFI setting to boot from the seconds slot, even after the conversion.)
My conversion was made possible by the fantastic work of github user badger707 who published a very well documented guide: https://github.com/badger707/m920q-dual-NVME.
In case you were wondering whether you could use the PCIe x8 slot to add two more NVMe SSDs, the answer is: “yes, but there are some caveats”. More information can be found in the following issue (and repo, badger707 as well): https://github.com/badger707/m920q-pcie-bifurcation/issues/1.
Ah, and if you consider buying a ThinkCentre Tiny you might want to check out the reference thread on ThinkCentre/ThinkStation Tiny on servethehome.com as well: https://forums.servethehome.com/index.php?threads/lenovo-thinkcentre-thinkstation-tiny-project-tinyminimicro-reference-thread.34925/.
*) The power consumption varies a lot depending on the use case, obviously. For my M920q with one SATA SSD and one NVMe SSD the momentary power consumption can go below 4W (fresh Proxmox VE installation, powertop applied, headless, turbo boost still enabled in UEFI), but on average it’s likely more in the range of 5-6W (all measured with a cheap power/energy meter, including the external power supply). With some VMs running (but idling), a HDMI dummy attached I would add 2-3W or so on top of that. As soon as the CPU has something significant to process, the power consumption will usually increase rapidly, and much more so than one would see with a Raspberry Pi 5. Then again, in my tests turbo boost was still enabled and I’d expect a significantly better performance from the ThinkCentre as well.
**) To name just two: Up to 64GB RAM in the ThinkCentre (officially supported 32GB) vs. 8GB max. in a Pi, many more PCIe lanes (Pi with only 1x externally with gen 2 speed by default, and only gen 3 if specifically enabled), or in other words: Much better upgradability, including for storage. This at a price that is surprisingly close to a Pi, if you factor in accessories that make both systems somewhat comparable performance-wise: ~170€ ThinkCentre M920q (Core i5-8500T, 16GB RAM, 256GB SSD, Power Supply) vs. ~150€ (Pi 5 8GB, Fan, Case, Power Supply, NVMe hat, 256GB NVMe). The ThinkCentre doesn’t have the Pi’s IO header, obviously… With passive cooling, the Pi can be absolutely quiet though. And not every application needs 8GB, nor a 256GB SSD. There are use cases for a Pi for sure, but for my applications I think the Pi is always a bit expensive for what I’m getting.