Comment by craftkiller

FWIW I run llama.cpp on AMD hardware using Vulkan. I've got no complaints but also nothing else to compare against.