Caixin Global – Latest China News & Headlines

Home >


CX Tech is Caixin Global's real-time tech news portal, featuring 24-hour news, short-form analysis, and roundups from business and tech media in China.

Huawei Launches ‘World’s Fastest AI Training Cluster’

By Zhao Runhua / Sep 18, 2019 01:16 PM / Business & Tech

Photo: VCG

Photo: VCG

Chinese telecoms giant Huawei unveiled Wednesday a homegrown artificial intelligence (AI) training cluster the company claims is the fastest in the world.

Called the “Atlas 900,” the cluster consists of around 1,000 Huawei Ascend 910 AI processors and has the computing power of half a million personal computers, Huawei said. It trained the much-lauded ResNet-50, a residual-network AI model often used in image recognition, in merely 59.8 seconds, Huawei Vice Chairman Hu Houkun said during the annual Huawei Connect conference in Shanghai.

The Atlas 900 has already been integrated into Huawei’s cloud infrastructure and is available for use, Hu said. Citing the research and advisory company Gartner, Hu said the computing sector’s market size is expected to reach $2 trillion within five years.

The development of the data and AI industries has created a need for technology companies to provide infrastructure that supports high computing power. Computer clusters fill that need by linking large numbers of computers into a single system and setting their nodes to perform the same task, resulting in higher speeds and increased capacity and scalability while reducing costs.

Huawei will continue to offer developers both hardware and software support, some of which will be open-source, Hu added. The company said at the conference that its Huawei Developer Program, which encourages developers to innovate on Huawei products and systems, will receive $1.5 billion to grow its number of participating developers from 1.3 million to 5 million.

Contact reporter Zhao Runhua (

Related: Experts: Huawei’s Offer to License 5G Technologies a ‘Smart PR Move’

Share this article
Open WeChat and scan the QR code