Page devoted to testing local ai models in the recent-ish Minisforum ms-s1 (year old, so sorta old in computer years). And maybe half-performance of a 96 GB Mac studio M5 ultra. The Minisforum (AMD 395) is running Windows for folks needing that, a couple grand cheaper and points there. So an interesting source for benchmarking local ai models on coding-adjacent workflows. Hoping Vulcan steps up its game with the new release to improve heap memory performance (looking at you VK_EXT_descriptor_heap)