Huawei unveils Ascend 960 supernode supporting 4,096 cards and NPO technology

On September 17, Huawei's HUAWEI CONNECT 2024 kicked off in Shanghai. During the event, the company launched the Ascend 960 Supernode, a new-generation AI computing infrastructure.

Key Specifications and Upgrades

• The maximum number of computing cards has been upgraded from 1,024 (previous generation) to 4,096.
• It is the first to adopt a key technology called NPO (Near-Package Optics).
• Designed to meet the training and inference demands of models with up to 10 trillion parameters.

Strategic Vision

"We are focusing on the computing power sector and will continue to innovate through system architecture to meet the explosive growth in domestic AI training and inference demands," said Wang Tao, Huawei's Deputy Chairman and Rotating Chairman.

Wang noted that the development of the Ascend 960 chip has exceeded expectations. The Ascend 960 air-cooled supernode and liquid-cooled supernode will be launched in the second and third quarters of next year, respectively.

Huawei's AI Strategy

Wang emphasized that the core of Huawei's AI strategy is computing power, with a focus on:

• Hardware monetization
• Building competitiveness through system architecture innovation around "Supernode + Cluster"
• Establishing an open-source and open computing ecosystem
• Supporting mainstream large models based on native Ascend training
• Actively embracing the "hundreds of models, thousands of applications" ecosystem

As of now, over 1,000 sets of the Ascend 910C Supernode have been deployed, and the Ascend 950 Supernode has already entered commercial use.

Breakthrough Technology: NPO and Hi-ONE

NPO is a critical technology for the Ascend 960 Supernode. Traditional large-scale computing systems rely heavily on copper interconnects, which face physical limits in transmission distance and bandwidth. NPO helps reduce interconnect power consumption and latency while solving cabling and heat dissipation challenges in high-density computing systems.

Huawei has developed a new-generation NPO-based optical interconnect product — Hi-ONE, which is:

• The first mass-produced NPO product in the industry
• The largest transmission capacity NPO product, reaching 7.2T
• The only NPO product with a built-in light source

In the Ascend 960 Supernode, 5,500 Hi-ONE modules replace the 48,000 800G optical modules that would otherwise be required. This results in:

• Power consumption reduction of over 550 kW
• Doubled system uptime without failure
• System availability reaching 99.8%

The product integrates "high bandwidth, high reliability" with "low latency, low power consumption," and together with the Lingqu high-speed interconnect protocol, forms a scalable supernode interconnect system.

Industry Perspectives

Zhu Keli, Founding Dean of the National Research Institute for New Economy and Chief Expert of the Intelligent Economy, noted that competition in computing power is not just about chips, but also about system architecture, software ecosystems, and overall industrial coordination — addressing pain points such as high communication latency, high costs, and insufficient reliability.

"Huawei's new-generation AI computing infrastructure provides a domestically producible foundation for large model training, driving upgrades in upstream and downstream industries such as optical communications and liquid cooling. At the same time, by opening interconnect standards, it helps reduce fragmentation in the domestic computing ecosystem and promotes the engineering deployment of domestic computing clusters," Zhu said.

Miao Fuyou, Chief Technology Officer of the Global Computing Consortium Secretariat, stated that China's biggest advantage in AI infrastructure lies in its full industry chain capability — covering chips, components, complete machines, power supply, and cooling systems — with rapid development across all areas. This is inseparable from China's manufacturing strengths and strong engineering capabilities. From a technical perspective, interconnect technology is a particularly strong area for China, playing a key role in the scalability and overall efficiency of supernodes and computing clusters.

Source: Economic Information Daily