Somewhere between "just buy a Raspberry Pi" and "build a rack server that hums like a jet engine" sits an entire universe of tiny AMD-powered boxes that can run circles around both. Beelink has become the default answer for "what should I run my homelab on," mostly because their lineup covers everything from "block some ads" to "run a local AI model" without you needing a second mortgage. We lined up four of them — SER3, SER5, SER5 MAX, and SER10 MAX — to figure out which tier actually matches what you're trying to build.
| Model | CPU | RAM | Storage | Best for |
|---|---|---|---|---|
| SER3 | Ryzen 3 3200U (2C/4T) | 8GB DDR4 | 256GB SSD | Single-purpose always-on tasks |
| SER5 | Ryzen 5 5500U (6C/12T) | 16GB DDR4 | 480GB SSD | Docker stack + light Proxmox |
| SER5 MAX | Ryzen 7 6800U/7735U (8C/16T) | 24GB LPDDR5 | 500GB SSD | Full Proxmox homelab |
| SER10 MAX | Ryzen AI 9 HX 470 (12C/24T) | 32GB DDR5 | 1TB SSD | Local LLMs + heavy virtualization |
SER3 — the "does one job forever" box
The SER3's Ryzen 3 3200U and 8GB of RAM aren't going to win any benchmarks, and that's fine — this tier isn't for benchmarks. It's for the boring, essential stuff that just needs to never turn off: a Pi-hole or AdGuard Home install that swallows every ad on your network, a scheduled backup job running through PoshRocket, or a lightweight file share for a household that isn't hoarding 4K Blu-ray rips. Idle power draw sits around 8W, so leaving it on 24/7 costs you pocket change over a year — and it's a licensed, clean Windows 11 Pro box out of the gate, though the hardware runs Linux fine if you'd rather go that route for server duty.
Don't expect it to transcode 4K Plex streams or juggle a stack of containers — independent testing shows this chip falling back to software decoding on high-resolution video, which is your first sign the GPU isn't built for that kind of work.
SER5 — the sweet spot for a real Docker stack
Bump up to the Ryzen 5 5500U and 16GB of RAM and something interesting happens: this stops being a novelty box and starts being an actual homelab machine. Reviewers running Proxmox on the SER5 report a typical stack — Home Assistant, Pi-hole, Nginx Proxy Manager, Plex, Portainer, plus 5-10 more containers — sitting comfortably under 30% CPU at idle with RAM to spare. The Vega 7 iGPU supports VA-API hardware transcoding on Linux, so Plex Pass users get real hardware-accelerated transcoding instead of the CPU sweating through every stream.
The 6-core/12-thread chip also has enough headroom for Proxmox itself, with AMD-Vi/IOMMU support making it straightforward to run Home Assistant OS as a VM alongside a Docker LXC for everything else. The one hard limit: it's got a single 2.5GbE NIC, which rules it out if you were hoping to run pfSense or OPNsense as your router — those want at least two network interfaces to separate WAN from LAN.
SER5 MAX — a whole lab in a lunchbox
Here's where it gets fun. Beelink sells the SER5 MAX with either the Ryzen 7 6800U or the newer-numbered 7735U — worth knowing before you compare listings by chip name alone, since independent testing confirms the 7735U is effectively a rebrand of the 6800U, same Zen 3+ architecture and near-identical performance. Don't pay extra chasing the newer model number; buy whichever is cheaper.
What actually matters is the 8 cores/16 threads and 24GB of LPDDR5, which is enough to run a genuinely complete lab: people have reported running Proxmox with a TrueNAS VM for file storage, a Docker host VM for containers, and a pfSense firewall VM — all at once, on one box the size of a paperback. That's a file server, a router, and a container platform living inside something that fits in a jacket pocket. Fan noise stays close to inaudible at idle and only picks up under sustained load, so a closet shelf is a fine home for it.
SER10 MAX — when "homelab" turns into "AI lab"
The SER10 MAX is a different animal: a Ryzen AI 9 HX 470 with 12 cores, 24 threads, a Radeon 890M iGPU, and an XDNA 2 NPU rated at 86 TOPS. This is the tier where "self-hosted" starts to mean local LLMs instead of just Docker containers. Reviewers report it comfortably running 13B parameter models and handling 27B models in Q4 quantization — genuinely useful for running your own coding assistant or chatbot without a cloud API bill or your prompts leaving the house. It's worth knowing its memory bandwidth (89.6 GB/s) trails purpose-built AI boxes like Strix Halo machines, so it's a strong prosumer local-AI rig, not a data center replacement.
Outside of AI workloads, it's simply the most homelab machine on this list — 10Gbps LAN, USB4, and enough CPU muscle to run Proxmox with several demanding VMs at once, with owners reporting it stays silent even after 5+ hours of sustained work.
So which one do you actually buy?
- You want a "set it and forget it" ad-blocker or backup box: SER3. Don't overspend on a job that barely uses 10% of a bigger chip.
- You want a real Docker/Home Assistant/Plex homelab without breaking the bank: SER5. This is the tier most people actually land on.
- You want Proxmox with a proper file server, router VM, and container host running simultaneously: SER5 MAX — grab whichever chip variant is cheaper that week.
- You want to run your own local AI models and a serious lab on top: SER10 MAX, and budget accordingly — it's the priciest tier by a wide margin.
Disclosure: This post contains affiliate links. If you buy through them, we may earn a small commission at no extra cost to you. Specs and performance notes are drawn from manufacturer listings and independent third-party reviews and forum reports, not our own hands-on testing of every unit listed here.