Particle.news

Download on the App Store

Huawei Open-Sources CANN Toolkit to Spur Ascend AI Ecosystem

The open source release of CANN is designed to accelerate AI innovation in China’s domestic chip ecosystem

Image
A Huawei Ascend AI chip is on display during a Huawei Data Infrastructure Summit at Fuzhou Digital China Convention and Exhibition Center on August 19, 2020 in Fuzhou, Fujian Province of China. (Photo by Zhang Bin/China News Service via Getty Images)
Image of Nvidia’s China-tailored H20 graphics processing units. Photo: Handout

Overview

  • Huawei announced at its Shenzhen developer conference that it will publish its Compute Architecture for Neural Networks toolkit under an open source license
  • The company aims to lower barriers for developers building applications on its Ascend AI processors by making CANN publicly accessible
  • Huawei has engaged leading AI firms, universities and research institutes to establish an open Ascend ecosystem around its domestic hardware stack
  • The open source pivot provides an alternative to Nvidia’s proprietary CUDA platform, whose license restricts use on non-Nvidia GPUs
  • The release follows Beijing authorities’ recent security scrutiny of Nvidia’s China-specific H20 GPU as part of a broader push for semiconductor self-reliance