A new chip creates a highly-efficient inference machine that scales from data center generative AI to edge computer vision applications.