Specs at a glance
| CPU | AMD Ryzen 9 7940HS — 8 cores / 16 threads |
| GPU / NPU | Radeon 780M integrated |
| RAM options | 16 / 32 / 64 GB DDR5 |
| Storage | NVMe SSD 1–2 TB |
| Power draw | 25–65 W |
| Form factor | 130 × 127 × 50 mm |
| Local LLM capability | Up to 13B Q4 |
| Agent score | 9/10 |
| Price point | €700–900 |
Overview
The UM790 Pro pushes the upper bound of the Linux mini PC tier. Ryzen 9 with Radeon 780M and 64 GB DDR5 ceiling brings Llama 3 8B Q4 to ~22 tok/s, Mistral Small 22B Q4 to ~10 tok/s. The 65W max TDP keeps thermals reasonable. Linux compatibility is solid (verified Debian, Ubuntu, Fedora). At €800, it's not the cheap option, but it's the highest-performance x86 mini PC we'd recommend for self-hosted AI in 2026.
Best for
- High-end Linux mini PC deployments
- Local LLM at 13B Q4 (Radeon 780M helps)
- Multi-tool agent workloads with browser automation
Not for
- Cost-optimised setups (cheaper options exist)
- Apple Silicon-equivalent unified-memory workloads
Compatible self-hosted agents
Tested working on Minisforum UM790 Pro (with the caveats from “Best for” / “Not for” above):
Hermes Agent
Safe default · Apache-2.0
OpenClaw
Specialist · MIT
Nanobot
Specialist · MIT
IronClaw
Specialist · Source-available
Moltworker
Specialist · Apache-2.0
See: all pocket AI hardware · edge AI hardware buyer's guide · how we test.