2026-08-22

If you've been eyeing local LLM inference without shelling out for a workstation GPU, the Bosgame M5 just got a proper review from ServeTheHome and it's worth paying attention to.

The M5 is a compact desktop built around AMD's Ryzen AI Max+ 395 — the chip with 16 Zen 5 cores, a beefy Radeon 8060S iGPU with 40 compute units, and crucially, 128GB of unified LPDDR5x memory. That shared memory pool is the headline for homelabbers. It means you can run large language models (70B+ parameter) entirely in memory without needing a discrete GPU with expensive VRAM. For context, an RTX 4090 gives you 24GB; this gives you five times that, accessible by both CPU and GPU.

ServeTheHome found the AI inference performance genuinely competitive for its form factor. It's not matching a multi-GPU server, but for a single-box local AI setup that fits on a desk and draws reasonable power, it's in a class of its own right now.

The caveats are real though. These Ryzen AI Max+ machines aren't cheap — expect north of £2,000–£2,500 depending on where you source one. Availability in the UK is still patchy; Bosgame ships internationally but you may face import duties. The memory is soldered and non-upgradeable, so 128GB is your ceiling. And while the iGPU memory bandwidth is decent, it's still well behind dedicated HBM solutions for pure throughput.

For homelabbers running Ollama, llama.cpp, or similar local inference stacks, this is arguably the most practical single-node solution available in a small form factor today. It also doubles as a perfectly capable daily driver or Proxmox host.

If you do pick one up and plan on cloning your existing boot drive across, an M.2 NVMe SSD enclosure makes the migration painless — or grab one from our shop: M.2 NVMe SSD Enclosure – USB-C 10Gbps with Magnetic Closure. Read the full review at ServeTheHome before committing — it's thorough.