Challenges in Architecting Next-Generation GPUs for Graphics and AI
Youngsok Kim is an assistant professor in the Department of Computer Science and the Department of Artificial Intelligence at Yonsei University. His research interests are computer architecture and systems with an emphasis on architectural support for next-generation heterogeneous system architectures which consist of CPUs/GPUs/NPUs/FPGAs. Before joining Yonsei University, he was a post-doctoral researcher at Seoul National University. He received his PhD and BSc degrees in Computer Science and Engineering from POSTECH. During his PhD studies, he worked as an intern at Google Mountain View office and Samsung Electronics’ System LSI Business.
In this talk, I will provide an overview of how GPU architectures have evolved over time to satisfy the high performance demands of graphics applications, how the GPU architectures turn out to be suitable for accelerating DNNs, and the challenges of the next-generation GPU architectures toward satisfying both graphics and DNN applications. I will first briefly discuss the characteristics of graphics applications and how modern GPU architectures are designed to efficiently accelerate the graphics applications. Then, I will explain how DNNs map well to the GPU architectures designed for graphics applications. After that, I will provide a few notable emerging applications which the next-generation GPU architectures should accelerate and how the characteristics of the emerging applications impose different performance requirements/challenges on GPU architecture designs.
ID : 897 821 7407
PW : 1nTQDY