High-speed optical interconnect for AIDistributed deep learning accelerator

Device Architecture Research Group

Technology

In distributed deep learning, how to connect multiple servers by high-speed communication greatly affects learning speed. We accelerate large-scale distributed deep learning by using high-speed AI optical interconnect powered with AI collective communication accelerator.

Features

Speed up data transmission with protocol specialized for AI data sharing
Perform AI data sharing computation during transmission to speed up learning

Application

By using this technology for the AI computing infrastructure, we can accelerate various corevo AI services

Application

Related technologies