Will AIGC Requires Disruptive Optical Interconnect Technologies

As one of the key enabling technologies for industry 4.0, Artificial Intelligence (AI) has been a hot topic and attracted lots of attention in past decades in both academia and industries. With the recent emergence of ChatGPT, generative AI (AIGC) has made headlines everywhere. The generative AI, which allows for the creation of the original content by learning from existing data with machines, may revolutionize industries and transform the way companies operate in the near future.

With unprecedented amounts of computing power required, AIGC may reshape the IT infrastructure as well. It has been shown that the total amount of computing power for large model training has been doubling every three to four months, much faster than Moore’s law. Massive parallelism has to be used to provide such high computing power due to the speed limit of a single GPU chip, generating huge demands for optical interconnects, which has been manifested by recent demand surge for optical transceivers in the market. 

Will AIGC requires significantly different optical interconnect technologies than current ones on form factors, bandwidth density, power consumption, latency et al? Can current optical interconnect technologies accommodate the AIGC demands? Will disruptive optical interconnect technologies be needed? This workshop will bring together experts from hyperscalers, AIGC service providers, optical interconnects.

Organizers: 

Chongjin Xie, Alibaba

Ning Liu, Soochow University

Ming Tang, Huazhong University of Science and Technology

Speakers:

Binzhang Fu, Alibaba

Jiansheng Feng, Tencent

QInfen Hao,  Institute of Computing Technology, Chinese Academy of Sciences 

Stefanos Dris, NVIDIA

Jiangwei Man, HiSilicon Optoelectronics

Brad Booth, Hyperphotonix

Ying Huang, Raintree Photonics

Gangxiang Shen, Soochow University

Weixing Gai, Peking University 

Diogo Costa, Ipronics 

Manish Mehta, Broadcom

Guangwei Hu, Nanyang Technological University