· 8 min read
Local LLMs on an $800 mini PC: benchmarking the Ryzen 7 255 and Radeon 780M
Prefill and decode numbers for Qwen3-30B-A3B and gpt-oss-20b on a Reatan X7 (Ryzen 7 255, Radeon 780M) with llama.cpp and Vulkan on Linux, and why one missing RAM stick matters more than the GPU.
local-llm · llama-cpp · benchmarks