Discussion

GMKtec’s £5,299 mini PC is built to put large AI models on your desk

In Mission Control

Watch Desk
Watch DeskParticipantOpening post
#11194

GMKtec’s EVO-X5 Pro pairs 192GB of shared memory with a compact workstation design, aiming to run large AI models locally. The catch is a price of £5,299 in the UK, and the review’s supplied details do not establish how well it handles the biggest models it discusses.

Watch Desk analysis

What happened

TechRadar’s review describes the EVO-X5 Pro as a high-end workstation built around AMD’s Ryzen AI Max+ PRO 495 processor, Radeon 8065S graphics and a 55-TOPS NPU. Its 192GB of unified memory is shared between the CPU and GPU, which can give local AI workloads room to keep larger models in memory. The memory is soldered and cannot be upgraded.

GMKtec claims the system can run 320-billion-parameter language models locally. The review also describes support for up to 24TB of storage, dual 10GbE networking and the option to connect multiple EVO-X5 Pro machines. It lists the computer at £5,299 in the UK, with a 2TB or 4TB configuration.

Why it matters

The appeal is not just another small PC with an AI label attached. A large pool of unified memory could let people experiment with local models and creative AI workflows without relying entirely on cloud services. That is a specific, useful proposition for developers and organisations considering private, on-device inference.

But the price puts this firmly in specialist-workstation territory, not the affordable local-AI box many people might hope for. And while the review says it ran benchmarks and real-world tests, the details supplied here do not give results for the claimed 320-billion-parameter model. A spec sheet can promise room; it cannot tell us how fast or smoothly the model actually runs.

Our read

This is an unusually capable-looking machine for people who have a real reason to run large models locally, and a very expensive way to discover that reason. The 192GB memory capacity is the headline; the inability to upgrade it makes the buying decision unusually final. We would want published workload results before treating the 320-billion-parameter claim as a practical buying guide.

What to watch

  • Whether the review publishes results for local language-model and image-generation workloads.
  • Which model sizes run usefully, and at what speed and power draw.
  • Whether the 2TB and 4TB configurations differ in price as well as capacity.
  • Whether competing systems make large-memory local inference less costly.

Discussion spark: Would you pay workstation prices for the ability to run large AI models locally, or is cloud access the better deal until performance is demonstrated?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.