Chinese tech giant Huawei has announced the launch of its fully open-source Ascend chip software ecosystem, aiming to support users in exploring the ecosystem's deep potential and conducting customized development independently.
Known as Compute Architecture for Neural Networks (CANN), the software serves as a bridge between high-level artificial intelligence (AI) training frameworks and Ascend chips, allowing users to access computing power without having to deal with chip-level complexity.
Huawei's announcement came at an Ascend computing industry conference held in Beijing on Tuesday (August 5th). Representatives from leading AI companies, universities, research institutions, and Huawei partners launched an initiative to develop the CANN open-source ecosystem. The initiative aims to combine industry strengths, explore the frontiers of AI, and build a thriving chip ecosystem.
Xu Zhijun, Huawei's rotating chairman, said that computing power remains at the core of the company's AI strategy.
Nearly every major company in the computing power card field has similar software. However, unlike the Compute Unified Device Architecture (CUDA), a closed-source ecosystem developed by US chipmaker Nvidia for nearly two decades, Huawei's CANN has taken a completely different path by adopting an open-source approach.
Huawei told Xinhua that after seven years of development, CANN has achieved significant breakthroughs in computing optimization, communication efficiency, and memory management. It is now able to provide comprehensive computing power support for all processes of AI model training and implementation.
The launch of CANN is Huawei's latest step in sharing its self-developed technology with the public. Previously, the company open-sourced its HarmonyOS operating system and MindSpore AI framework.
