Shanghai, China – Huawei has announced the Peerium Computing Architecture, a system designed to connect processors at the million scale and enable them to operate as a single computer to meet growing demand for AI computing capacity.
The architecture uses nested parallelism, unified memory addressing and peer interconnect technology to scale computing across large numbers of processors. Huawei said it introduces Nested Bulk Synchronous Parallel (Nested BSP), extends the traditional von Neumann single-machine architecture, and moves away from the master–slave model widely used in computing systems.
Unified interconnect for compute, storage and networking
Huawei said its UnifiedBus (UB) technology is a key component of the architecture. Based on a single open protocol, the high-speed interconnect connects CPUs, NPUs, memory, solid-state drives, network interface cards and switches.
The company said UB enables peer-to-peer connections across computing, storage and networking resources, supporting the architecture’s ability to scale.
Atlas systems built on Peerium
The Atlas 950 SuperPoD and SuperPoD-based SuperClusters are the first-generation products built on the Peerium Computing Architecture, according to Huawei.
The company said an Atlas 950 SuperCluster with 256,000 cards is already being deployed. Its Atlas 960 system, which uses near-packaged optics (NPO), is currently undergoing testing.
“In the AI era, Huawei is drawing on the Peerium Computing Architecture we pioneered to continuously build the SuperPoDs and SuperPod-based SuperClusters that meet customer needs for training and inference, making computing power available everywhere and intelligence accessible to all,” said Eric Xu, Rotating Chairman at Huawei.
