AMD aims to enhance the energy efficiency of its chips by 20-fold by 2030, leveraging rack-scale architectures to achieve this goal. Sam Naffziger, AMD’s SVP, highlights that larger devices often demonstrate greater efficiency, enabling more computing power in a single package. The company’s innovative chiplet architecture has strengthened performance per watt, culminating in the MI300 series of APUs and GPUs, which features advanced 3D stacking technologies. Looking ahead, AMD plans to release its first rack-scale compute platform, the MI400, next year, following a model similar to Nvidia’s GB200 NVL72. Anticipating advancements in photonic interconnects, AMD envisions replacing copper to address power and bandwidth demands. Moreover, the company recognizes the importance of hardware-software co-design to optimize performance for AI applications, while also acknowledging the challenges of rapidly evolving AI technologies. Overall, AMD is poised to push the frontiers of energy efficiency and performance in data centers.
Source link
Share
Read more