GMKtec EVO-X2

What owners say

This is a local-AI box first. With 128 GB a 70B model at Q6 fits entirely in GPU memory, but at under 4 tok/s it crawls; where the machine flies is on mixture-of-experts models, with gpt-oss-120B between 34 and 46 tok/s and Qwen3-Coder-30B-A3B between 50 and 71, while dense 32B models sit near 10. How much memory the GPU gets depends on the operating system: Windows caps it at 48 GB on the 64 GB model and 96 GB on the 128 GB one, while Linux reaches about 120 GB. It idles quiet at 27 dB(A) and becomes clearly audible under load at 47 dB(A); the fan curve is coarse and there is a P-mode button. Reviewers measured cool processors, but at least one owner saw 95-100 °C while gaming until he repasted, and the stock NVMe heatsink is widely disliked. Support is the sore point: opaque BIOS updates, machine-translated replies and months without a published security patch. It must stand vertical, has no VESA mount, runs off a 230 W external brick and offers only 2.5GbE at this price.

For whom

People who run local language models, especially mixture-of-experts ones, and want them in a single quiet-ish box.

Not for

Gaming on a budget, where a tower gives more frames per dollar, or anyone who needs ROCm on Linux to work without a fight.

Our own summary of 8 published sources, reviewed Sep 17, 2026. We have not tested this Mini PC ourselves; the sources are listed under About this data.