“NVIDIA's Grace Blackwell architecture with NVLink 72 delivers 35 times higher performance per watt for AI inference compared to the previous generation.”