Top picks by use case
| Use case | Recommended | Why |
|---|---|---|
| Low-power always-on (DNS, *arr, Pi-hole) | Intel N100 / N150 mini PC | ~6-10W idle, plenty for 10-15 light containers |
| Proxmox / virtualization with 2.5GbE+ | Beelink EQ / SER mid-range | 32-64GB RAM ceiling, 2.5GbE standard |
| Cluster node with 10GbE + dual NVMe | Minisforum MS-01 | SFP+ 10GbE, two NVMe, full PCIe slot |
| Plex / Jellyfin with hardware transcoding | Any 12th-gen+ Intel mini PC | Quick Sync handles multiple 4K transcodes |
| AI / LLM experimentation | AMD Ryzen AI mini PC with 64GB+ | iGPU/NPU can run 7-14B parameter models |
The main contenders
The mini PC that ate the homelab world. Intel Core i9/i7 with 96GB RAM ceiling, dual 10GbE SFP+, dual 2.5GbE, three NVMe slots (one PCIe Gen4 x4), and a usable PCIe x16 slot. It's the only mini PC that competes with full SFF towers for cluster nodes — and it idles around 15-20W.
Best for: three-node Proxmox cluster, anyone wanting 10GbE without a separate switch port, serious self-hosters.
Under $200, four cores, 16-32GB RAM, dual 2.5GbE on the better SKUs. Idle power is 6-10W. Perfect for the "always-on services" tier: Pi-hole, Home Assistant, *arr stack, Vaultwarden, Tailscale, Uptime Kuma. Quick Sync handles a single 4K transcode.
Best for: first homelab, headless server replacing a Raspberry Pi, dedicated DNS/router-adjacent box.
Ryzen 7 or Core i7, 32-64GB RAM ceiling, dual 2.5GbE, two NVMe slots. The "next step up from N100" for people running Proxmox with a handful of VMs. Cheaper than MS-01 if you don't need 10GbE.
Best for: Proxmox / Docker host, modest VM workloads, second cluster node.
Newer Ryzen AI mini PCs ship with 64-128GB unified memory and integrated graphics capable enough to run 14-30B parameter LLMs in 4-bit quantization. Power draw is higher than N100 but still under 50W under load.
Best for: local LLM experimentation, anyone wanting a single box that handles homelab + AI.
The "TinyMiniMicro" niche — refurbished enterprise SFFs with 6th-10th-gen Intel CPUs. 16-32GB RAM, sometimes a single NVMe slot, gigabit NIC. Often under $100 used. Slower than new mini PCs but excellent value for a learning lab.
Best for: first homelab on a budget, second/third node for cluster experiments, learning Proxmox without spending much.
What actually matters
- RAM ceiling — most important. 16GB fills up fast once you run VMs. 32GB is the practical floor; 64GB+ is comfortable.
- NIC speed — 1GbE caps your between-node traffic. 2.5GbE is the new minimum; 10GbE is the cluster sweet spot for shared storage.
- NVMe slots — two is meaningfully better than one (mirror, or boot + data separation).
- CPU generation — newer Intel iGPUs have far better video transcoding. Newer AMDs have better power efficiency under load.
- Idle power — a box that runs 24/7 at 15W extra costs ~$15-30/year more than one at 5W. Over 5 years, this adds up.
What doesn't matter as much
- "i7 vs i9" — for homelab workloads (mostly idle), the cheaper SKU usually wins on price/performance.
- Wi-Fi — homelab servers should be wired. Wi-Fi is a check-box feature you'll never use.
- Fancy GPUs — most homelab workloads don't need discrete GPU. iGPU + Quick Sync covers Plex transcoding.
Cluster vs single node
One MS-01 with 64GB RAM beats three N100s with 16GB each for most workloads. Clusters become interesting once you care about high availability, live migration, or want to learn Kubernetes at small scale. See homelab cluster design for the trade-offs.
Frequently Asked Questions
How much RAM do I need for a homelab?
16GB is the floor for a basic Docker host. 32GB is comfortable for Proxmox with 5-10 VMs. 64GB+ gives headroom for AI workloads, larger VMs, and not having to think about memory pressure.
Can I just use a Raspberry Pi?
For very light workloads (Pi-hole, Home Assistant, Tailscale exit node) — yes. For Docker with 10+ containers, virtualization, or anything storage-heavy, a mini PC is far better value once you factor in SSD storage and reliability.
Is the MS-01 worth the premium?
If you want 10GbE without a separate add-in card, three NVMe slots, and a real PCIe slot — yes. If you're running a single node with 2.5GbE, a cheaper mini PC saves money for the same workload.
How loud are these?
N100-class mini PCs are near-silent. Higher-end mini PCs (MS-01, Ryzen AI) spin fans audibly under sustained load but are quiet at idle. None are as loud as used 1U enterprise servers.
What about ARM mini PCs?
ARM mini PCs (Rockchip RK3588 boxes) are improving but still trail x86 in software ecosystem — many homelab apps don't have first-class ARM container images. Apple Silicon Macs are excellent ARM servers but proprietary. For most homelabs, x86 mini PCs remain the better choice.