Comparing the Asus Ascent GX10 and ExpertCenter Pro ET900 NG3: Which LLMs Can Each Run?

A side-by-side look at two compact AI machines — the 128 GB Asus Ascent GX10 and the 748 GB Asus ExpertCenter Pro ET900 NG3 — and the size of language models each one can actually run. Models from 4 billion up to 397 billion parameters are put through the same test so you can see how token throughput changes as the workload grows.

■ Hardware Overview
- Design and features of the GX10, including its cooling, power indicator, and stackable magnets
- Positioning of the ET900 NG3 as a workstation-class counterpart

■ Model Size Tests
- A 4 billion parameter model
- A 27 billion parameter dense model
- Models around 120 GB that approach the GX10's memory limit
- A 142 GB model that only fits on the larger machine
- A 397 billion parameter mixture-of-experts model

■ Takeaways
- How measured tokens per second compare between the two machines at each size
- Where each machine's memory capacity becomes the deciding factor

Anyone choosing between a compact AI box and a larger workstation will come away with a clear sense of which model sizes suit which machine, and what to weigh up before deciding.