I've been seriously considering building out a home lab using a modern mini PC instead of a traditional tower setup. With the latest generation of processors from AMD Ryzen and Intel, plus the ability to upgrade RAM and NVMe storage, they are a great choice I think. But I'm curious about real-world performance when running multiple VMs or containerized workloads?
Here's what I'm weighing on purchasing. Either a Minisforum, Beelink, or Geekom mini PC with a Ryzen 7 or high-end Intel processor. I want something that could handle Proxmox, Docker, and Kubernetes for self-hosting projects. I'm wondering about practical limitations. Has anyone pushed a mini PC to its limits with heavy virtualization? Are there thermal throttling issues when running sustained workloads? How does NVMe storage expansion work on these devices, and is it truly user-friendly?
I'm also curious about the GPU angle. Some of these mini PCs support external GPUs or have integrated graphics capable enough for local LLM inference. Are any of you running AI workloads or experimenting with local language models on mini PC hardware? What's your experience been with stability and performance?
The other consideration is memory upgrades. Most mini PCs max out at 32GB or 64GB depending on the model. Is that enough for a serious home lab setup, or are you finding yourselves limited? What's your sweet spot for RAM when running multiple services?
Keegan,
I've switched almost entirely to mini PCs for my home lab. For Proxmox, Docker, and Kubernetes, they're way more capable than most people might think and I think the biggest limitations isn't CPU, it's RAM. I'd look for a model that supports at least 64GB, preferably 96GB+. Most of the vendors didn't really publicly say they supported 128 GB with the new 64 GB SODIMMs but I didn't find any of the mini PCs that I tried that failed to support 128 GB. But that is just a side point. Especially with how expensive RAM is these days, it probably isn't feasible to max one out any way.
As for thermals, I've never had issues with my Ryzen-based systems under normal virtualization workloads. Also, my MS-01s are trucking away and doing fine with normal workloads as well, no thermal issues. For AI, integrated graphics are fine for experimenting, but if local LLMs are a major goal, you'll want a dedicated GPU, which you can do with an OCuLink card, etc. Some mini PCs have a slot that will fit a low-profile GPU though too.
Overall, I'd take a modern Ryzen mini PC over a traditional tower for most home lab use cases today due to their uniform cores and they are just great performers.
Im just using recycle worthy computers. One is a Dell optiplex running at 4ghz, 32Gb of DDR3 1600 RAM. The other is an Intel server board with Xeon 3.2Ghz, 16Gb RDDR3. They do really good for what they are. I have noticed i can play shows with no issue but movies will not play without buffering. I will eventually get a GPU to help with transcoding. That will be the more expensive purchase that is worth the splurge.
@roblewis That is awesome though and a testament to what you can still do with second hand hardware. Are you using Plex or Jellyfin by chance?
@brandon-lee Jellyfin. I set it up following Automation Avenue’s New ARR stack video. It runs in docker in the LXC.
@roblewis very cool!