Radeon 780M iGPU as budget solution for MoE LLMs

Reddit tip highlights the Radeon 780M iGPU for budget local inference, using partial expert offloading on MoE models like Qwen 3.6 35B-A3B Q8 MTP to boost token generation while keeping costs low.