It is now imperative for computer architects to develop AI-optimal computer systems. Therefore, it is becoming extremely difficult to architect fast, scalable and cost-effective server architectures for a specific application. To address the challenges, I will first describe the potentials and challenges to develop AI-optimal server architectures. Next, I will introduce our recent research outcomes to target various challenges in architecting AI-optimal computer systems: device management [MICRO 2015/ISCA 2018], multi-AI inference [ISCA 2020], large-scale training [MICRO 2020] and virtual machine [OSDI 2020].
Jangwoo Kim is a full professor in the Department of Electrical and Computer Engineering at Seoul National University. He earned his PhD degree from Carnegie Mellon University, and his BS and Master degrees from Cornell University. Prior to the academic career, he contributed to architect and develop UltraSPARC T4 CPUs and servers at Sun Microsystems and Oracle Corporation. His current research interests lie in server and storage architecture, FPGA-assisted computing, cryogenic computing, and AI/neuromorphic computing.