AMD just proved you don't need a graphics card to run a serious AI model at home
Day-zero support for the new Qwen3.8 27B just landed on two AMD boxes — one needs a graphics card, one doesn't — and the tokens-per-second numbers…
Aliteq
Lena Fischer · AI & Local Compute Editor
Qwen3.8 27B, a new dense model from Alibaba's Qwen team, launched August 14, 2026, needing roughly 24GB of memory to run
AMD's Radeon AI PRO R9700 (32GB, $1,299) hits 51.8 tokens/sec on the model using Vulkan backend optimizations
The Ryzen AI Max+ 395 — integrated graphics only, no discrete card — hits 24.5 tokens/sec on the same model
Both run through free tools: LM Studio for a GUI, or AMD's Lemonade platform for app integration
The catch: this runs on Vulkan/ROCm, not CUDA, and tooling maturity still trails Nvidia in places
Buy call
The R9700 wins on price-per-token if you already have a PC. The Ryzen AI Max mini PC wins if you want a single silent appliance that might scale to bigger models later.
Aliteq
Read the full story
AMD just proved you don't need a graphics card to run a serious AI model at home