Strix Halo local LLM guide: 63-97 t/s direct MoE on Ryzen AI MAX+ 395 / 128GB unified memory. Setup, model choices, benchmarks, and raw evidence.
benchmark amd optimization vulkan inference rocm mini-pc unified-memory beelink llm llama-cpp local-llm ollama gguf rdna3 ai-pc strix-halo gfx1151 dgx-spark ryzen-ai-max
-
Updated
May 17, 2026 - Python