GMKtec EVO-X2

What it is

GMKtec’s current Ryzen AI Max+ 395 mini PC. 16 Zen 5 cores, Radeon 8060S (40 RDNA 3.5 CUs), 64GB or 128GB of onboard LPDDR5X-8000, dual M.2 2280, Windows 11 Pro. Same Strix Halo silicon as the Framework Desktop, in a sealed 193 × 186 × 77 mm box that is meant to stand vertical. This page is the 128GB machine.

Who it is for

People who want a 24B–70B local model and a real Windows PC on the desk, already assembled, and can live with llama.cpp / ROCm / Vulkan. Wrong if your tools assume CUDA, if you want to open the box in two years, or if you thought $2,200 bought 128GB.

What it actually runs

• Comfortable: 24B–32B dense at usable speed on the iGPU.
• Possible: 70B-class dense and 120B-class MoE if you buy 128GB and accept that 256 GB/s is not Mac Studio bandwidth. Capacity is the win. Decode is not.
• Not the win: training, CUDA-only stacks, or beating a 5090 on anything that already fits in 32GB VRAM. Ignore vendor token-rate slides.

Standout details

• Price snapshot 30 Aug 2026, GMKtec US store JSON: 64GB + 1TB $2,199.99; 128GB + 1TB $3,499.99; 128GB + 2TB $3,649.99. Those 128GB SKUs were listed available. The “from $2,199” number is the 64GB config.
• Memory is soldered. Pick 128GB now or do not buy this for large models.
• 256-bit LPDDR5X-8000, ~256 GB/s theoretical. Same limiter as Framework.
• 2.5GbE, Wi-Fi 7, dual USB4, HDMI 2.1 + DisplayPort. No 10GbE. One-year warranty.
• Reviews and owners keep coming back to the same chassis problem: a ~120W (140W peak) chip in a 77 mm box. Performance mode is loud. The feet say vertical.
• GMKtec also sells an EVO-X3 with the same chip, OCuLink, and a taller price (~$3,800). Same memory ceiling. This listing is the X2.

Honest verdict

Buy the 128GB EVO-X2 if you want Strix Halo already built and you can find that SKU near $3,500. Skip the 64GB config for this job. If you want to repair the machine, get Framework Desktop. If you want CUDA, shop Spark / GX10.