I've been diving into the world of mini PCs for home lab setups, and I'm curious about the best approach for running multiple services. Right now, I'm torn between setting up Proxmox for full virtualization or going the Docker route for containerization. Both seem viable on modern mini PCs like the Minisforum or Beelink models with decent specs.
The appeal of Proxmox is having isolated VMs for different services—it feels safer and more flexible. But Docker seems lighter on resources and faster to deploy. I'm wondering: what's your experience been? Are you running a hypervisor setup on your mini PC, or are you using containers? What hardware are you working with, and how's the performance been?
I'm particularly interested in:
- Real-world performance comparisons on mini PCs with 16GB or 32GB RAM
- NVMe storage considerations—how much do you need for multiple VMs or containers?
- Whether GPU acceleration matters for your workloads
- Tips for managing power consumption while running 24/7
- Any gotchas you've hit with specific mini PC models
I'm also curious if anyone's experimenting with AI workloads or local LLMs on their mini PC home labs. Are you finding the hardware limiting, or are mini PCs surprisingly capable for these tasks? Let's share our setups and learn from each other's experiences!
Leonard,
Glad to have you here in the forums. I think mini PCs are incredibly powerful these days. You can get some really great hardware in a small package. I am currently running my Proxmox mini lab on Minisforum MS-01s (5 of them) and also running Ceph software-defined storage for shared storage between them for VMs and Kubernetes persistent volumes.
I think it is really hard to say it will be ALL one or the other (VMs vs containers). I have mostly Docker container hosts running as virtual machines. Then have lots of containers running in these environments. Also, I have VMs running as Kubernetes nodes for control and worker nodes. I like to run a hypervisor on my mini PC hardware so that I have things like live migration and easy backups of my data. If you are running more bare metal configurations with Docker containers, it limits you in certain ways. I definitely like to have that abstraction layer there in between the VM and the containers running.
Also, you can play around with disabling "Core Performance Boost" with AMD or similar with Intel as this will limit the power draw of the CPU and limit the frequency bursts that are normal with modern CPUs. This will definitely limit performance much lower than the CPU is capable of, but it will keep things running cooler and much less power draw in my experience.
Also, you can definitely run LLMs with mini PCs. You can do an OcuLink port on a mini PC to link up to an eGPU if you want or some mini PCs like the MS-01 and MS-02 Ultra can handle certain low profile performance video cards. Hopefully, these points will help you with some of the decision making.