• Technical Conference:  30 March – 03 April 2025
  • Exhibition: 01 – 03 April 2025
  • Moscone Center, San Francisco, California, USA

Low-Latency High-Speed Optical Interconnection Technologies for AI Compute Era

Thursday, 28 March, 11:30 – 12:30

Theater II

Session Description:

For AI deep learning models, the vast amount of high-performance computing power requires a transmission network that is key to computing. The computing workload is allocated to thousands of GPUs in a cluster to train a large language model. This calls for a high-throughput and low-latency network collaborating with many chips to integrate their vast computing power.

New technologies that reduce transmission latency include simplified network structures such as the Ethernet's three-layer CLOS architecture (spine-leaf) and new Ethernet protocols such as RoCEv2. So, what new technologies can reduce transmission latency in optoelectronic devices? This seminar will discuss next-generation optical technologies (such as linear drive and CPO) that can reduce transmission latency. We will also discuss the development trend and technological innovation direction of AI foundation models, new computing power, and next-generation low-latency and high-speed optical connection infrastructure, which will help to build an intelligent and efficient optical interconnection world in the future.

Organized by:


Frank Chang, Source Photonics, United States


John D'Ambrosia, Futurewei, United States

Jim Theodoras, HG Genuine, United States

Jianying Zhou, Hisense Broadband, United States

Chen Zhu, Baidu, China