AI China

Huawei Announces New AI Computing Strategy & Unveils World’s First 7nm ARM Data Centre Chip

At the Huawei Intelligent Computing Conference yesterday in Beijing, the Chinese tech giant announced a new intelligent computing strategy for enterprises, launched the world's first 7nm ARM-based data centre processor, and introduced a new-generation intelligent computing platform.

At the Huawei Intelligent Computing Conference yesterday in Beijing, the Chinese tech giant announced a new intelligent computing strategy for enterprises, launched the world’s first 7nm ARM-based data centre processor, and introduced a new-generation intelligent computing platform.

New Intelligent Computing Business Unit and New Strategy

In a move reflecting the company’s ambitions in AI and cloud computing, Huawei upgraded its Server Product Line to an Intelligent Computing Business Unit. Company President Long Qiu said the new unit will focus on overcoming four major challenges in the computing industry:

  • Develop chips and innovative technologies to provide abundant and affordable computing power
  • Adapt to the industry’s rapidly changing environment through universal engineering capabilities
  • Establish cloud-edge collaborative architecture and high-bandwidth, low-latency, seamless network coverage for data collaboration and interoperability
  • Address the AI talent shortage with an all-in-one solution to make the tech “as simple and convenient as using hydro”

Huawei’s new strategy will provide intelligent servers for traditional business, offer cloud services for digital transformation, and drive enterprise development by building intelligence with heterogeneous/edge computing.

New 7nm ARM chip

Huawei also officially announced the world’s first 7nm ARM-based data centre processor, which will empower the company’s next-generation “TaiShan” servers. The Hi1620 is built on ARM v8 architecture cores with the memory of eight DDR4-3200 channels per socket and support for PCIe 4.0 and CCIX. The new chip increases bandwidth by 33 percent and efficiency by 20 percent compared to its predecessor Hi1616. Qiu revealed that the Hi1620 is already being used by Huawei internally.

The Hi1620 is the latest of the company’s recent chip rollouts. The Ascend 910 and Ascend 310 7nm-based AI chip were introduced two months ago at the Huawei Connect 2018 Conference. The chips run on the cloud for training and inferencing and are built on Huawei’s homegrown Da Vinci architecture, which features scalable memory, compute, and on-chip interconnection.

Huawei’s chip portfolio also includes the Atlas 200 AI accelerator module; the DC-facing Atlas 300 AI accelerator card; the edge-oriented Atlas 500 AI edge station; and the one-stop AI platform Atlas 800 AI appliance for enterprises. These chips will be packed into Huawei’s upcoming Atlas intelligent computing platform — release scheduled for March 2019 — to meet computing demands for devices ranging from cloud to edge to mobile.

Also announced today were Huawei’s first intelligent management chip Hi1711, which provides intelligent fault management capabilities; a 16-nm intelligent SSD controller processor; and the company’s third-generation intelligent network chip Hi1822.

Huawei also pledged CN¥1 billion (about US$150 million) to nurture AI talents and developers in a joint effort with Chinese elite universities.


Journalist: Fangyu Cai, Tony Peng | Editor: Michael Sarazen

%d bloggers like this: