Huawei Unveils Peerium Architecture to Unite Million-Scale Processors as Single Computer
Huawei announces Peerium Computing Architecture designed to enable processors at million scale to operate as one unified computer.
AI-generated summary. This news item was written automatically by an AI pipeline from published reporting and was not reviewed by a person before publication. It may contain errors. Read the original source.
New Peerium Architecture Targets Massive-Scale Computing
Huawei has announced its new Peerium Computing Architecture, designed to enable processors at the million scale to work as one computer. The architecture represents a significant step forward in handling the exponential growth of AI workloads and infrastructure demands.
Foundation Models Approaching 10 Trillion Parameters
Huawei calculates that foundation model parameters are rapidly approaching 10 trillion, with projections that foundation model parameters will exceed 100 trillion by 2030. This trajectory underscores the scale of computational resources needed to support next-generation AI systems.
Inference Token Demand Surging in China
Daily inference token consumption in China has surged to around 500 trillion tokens. Huawei projects that daily inference token consumption in China will reach quintillions by 2030, highlighting the massive infrastructure requirements needed to support AI services at scale.
On-Device AI Parameters Expand Tenfold
On-device AI parameters for smartphones have expanded significantly, growing from 3 billion in 2024 to 30 billion by the end of 2026. This tenfold increase demonstrates the rapid acceleration of AI capabilities being pushed to edge devices.
UnifiedBus Interconnect Technology
Huawei’s UnifiedBus (UB) interconnect technology is built on a single open protocol and is designed to scale without limit to connect CPUs, NPUs, memory, SSDs, NICs and switches. The technology aims to provide seamless connectivity across diverse hardware components at unprecedented scale.
Atlas SuperCluster Deployments Underway
An Atlas 950 SuperCluster with 256,000 cards is already being deployed. Complementing this infrastructure, the Atlas 960 system, based on Near-Packaged Optics (NPO), is currently under testing.
Hi-ONE Optical Engine Achieves 7.2 Tbps Throughput
Huawei’s Hi-ONE optical-interconnect-node engine achieves a transmission capacity of 7.2 Tbps per single engine. Huawei describes Hi-ONE as the industry’s first NPO product ready for mass production and the first NPO product with a built-in light source.
Source: Computer Weekly