Unified Memory, Explained: Why Mini PCs Can Run 70B Models a Big GPU Can't

# Summary Mini PCs with unified memory architecture can run large AI models like 70-billion-parameter models that discrete GPUs cannot fit, because they share a single large memory pool (up to 128GB) rather than being limited by separate VRAM. However, they achieve this capacity advantage at the cost of significantly slower inference speed due to lower memory bandwidth compared to high-end GPUs like the RTX 5090. The trade-off between capacity and performance explains why mini PCs are emerging as a viable option for running local large language models despite their computational limitations.