Key facts
- Shanghai Biren Intelligent Technology has unveiled a new architecture capable of interconnecting up to 1,024 GPUs.
- The solution utilizes advanced optical connections and near-packaged optics (NPO).
- It integrates optical engines directly onto the company's upcoming BR2xx series boards.
- The design separates computing and networking nodes, allowing for standard server deployments.
- This approach aims to eliminate the need for customized racks and shorten product development cycles.
Hong Kong-listed chipmaker Shanghai Biren Intelligent Technology Co. has unveiled a next-generation architecture capable of interconnecting up to 1,024 graphics processing units (GPUs), adopting advanced optical connections to power massive artificial intelligence workloads.
The solution, showcased at the World Artificial Intelligence Conference, integrates optical engines directly onto the company’s upcoming BR2xx series boards using near-packaged optics (NPO). By physically separating computing and networking nodes, the design allows for standard server deployments — eliminating the need for customized racks and drastically shortening product development cycles.