Interview: „Efficiency has to be a central design principle for AI data centers“

Huge data centers with enormous AI power are emerging everywhere. In this interview with c't, AMD's Dr. Zacharia explains how their efficiency can be improved.

Gift this article
listen Print view

(Image: Oak Ridge National Laboratories)

15 min. read
Contents

Artificial intelligence on an industrial scale requires enormous computing power and storage capacity, which only specialized data centers can provide. Despite the exorbitant costs, dozens, if not hundreds, of such data centers are currently being built. This causes concern for many observers not only because of the dominance of a few large US corporations. Ecological issues also play a major role as an end to the AI boom is hardly in sight and with the large number of data centers,their global ecological relevance is increasing.

To increase the efficiency of AI data centers, AI accelerators, memory and network technology, cooling, and server housings would have to be developed jointly and optimally coordinated with each other. We spoke with Dr. Thomas Zacharia about the specifics of AI data centers and their requirements for infrastructure and energy. Since March 2024, he has been Senior Vice President for Strategic Technical Partnerships and Public Policy at AMD, responsible worldwide for partnerships with governments, research institutions, and industry to promote AI and high-performance computing.

Before that, he was director of the Oak Ridge National Laboratory (ORNL) and led four systems that held the top spot on the TOP500 supercomputer list. He is also chairman of the board of the Federal Reserve Bank in Nashville.

Das war die Leseprobe unseres heise-Plus-Artikels "Interview: „Efficiency has to be a central design principle for AI data centers“". Mit einem heise-Plus-Abo können Sie den ganzen Artikel lesen.