Huawei's David Wang: Accelerate Intelligence with Custom-built AI Models for All Industries

[Shanghai, China, September 20, 2023] During HUAWEI CONNECT 2023, David Wang, Huawei's Executive Director of the Board, Chairman of ICT Infrastructure Managing Board, and President of the Enterprise BG, delivered a keynote speech titled "Accelerate Intelligence". Wang shared Huawei's views on and experience in helping industries make the most of intelligence. He also announced the launch of Huawei's new Atlas 900 SuperCluster. This new AI computing cluster, the latest offering in Huawei's Ascend series of computing products, makes use of a brand-new architecture. In addition, Wang announced nine new intelligent industry solutions based on Huawei's Intelligent Transformation reference architecture. These solutions are designed to meet the specific needs of different industries, including finance, government, manufacturing, electric power, and railway.

"A new chapter of intelligent transformation is unfolding," David Wang said. "We now stand at the threshold of a new and intelligent world, with vast opportunities and challenges ahead of us. We need to work together, dig deep into industry-specific scenarios, and build a solid computing backbone to power countless new AI models and applications. Together, we can help all industries go intelligent, and help them do it faster."

The challenges keeping industries from going intelligent

With many recent breakthroughs in foundation models, a vast range of new AI models and applications are emerging. AI is also being deeply integrated into industries, serving more and more industry scenarios. However, data, computing power, algorithms, and application deployment are struggling to keep up and will be critical to enable the intelligent transformation of industries.

To address these challenges, Wang called for joint efforts to support intelligent connectivity, intelligent computing, and intelligent industries, which will be key to addressing the ongoing problems with AI implementation and scenario-specific models. He explained that this is how different industries can make most of intelligence.

The Ascend Atlas 900 SuperCluster

Massive computing power is critical for the development of foundation models. By moving away from traditional server stacking approaches, and supported by recent innovations, Huawei's newest AI clusters stand out at both the system and architecture levels. In addition, through the integration of computing power with transmission and storage it has been able to overcome the current bottlenecks in computing power.

More and more foundation models trained with trillions of parameters are emerging, and so Huawei launched their newest Atlas 900 SuperCluster, specifically designed for training massive AI foundation models. The Atlas 900 SuperCluster comes packed with Huawei's state-of-the-art Xinghe Network CloudEngine XH16800 switch featuring high-density 800GE ports.

All told, the cluster's innovative super node architecture greatly boosts the cluster's overall computing power and takes the speed and efficiency of foundation model training to an entirely new level.

In addition, Huawei has leveraged its strengths in computing, storage, network, and energy to systematically improve system reliability at the component, node, cluster, and service levels. System reliability is incredibly important for training massive foundation models, and this approach has effectively extended the cluster's ability to support continuous model training from several days to a month or more.

Having the right hardware is the key to unleashing massive computing power and developing foundation models. To further speed up foundation model development, Huawei also launched a more open and easier-to-use Compute Architecture for Neural Networks (CANN) 7.0. This architecture is not just compatible with other available AI frameworks, acceleration libraries, and mainstream foundation models, but also opens up more lower-level capabilities. More open capabilities ensure that AI frameworks and acceleration libraries can directly invoke and manage computing resources, so that developers can customize their own high-performance operators to make their foundation models more unique and competitive.

Huawei has also upgraded its Ascend C programming language for Transformer network models. The more efficient programming and simplified operator implementation logic reduce the development time for a fusion operator from two person-months to two person-weeks, dramatically accelerating AI model and app development.

Working with customers and partners to launch nine intelligent industry solutions

Over the past three years, Huawei has established a flock of integrated teams. They dive deep into industries and scenarios, bringing specialized groups of experts closer to customer challenges and more tightly incorporating horizontal R&D resources to help industries go intelligent. This operational model has already helped Huawei more closely work with partners to create over 200 dedicated intelligent transformation solutions for more than 20 industries, including urban governance, finance, transportation, and manufacturing. These solutions have already been implemented in many real-world projects.

During his keynote, Wang launched nine new intelligent industry solutions based on Huawei's Intelligent Transformation reference architecture. A joint effort between Huawei, its customers and partners, these solutions are designed to meet the specific needs of different industries, including finance, government, manufacturing, electric power, and railways. Huawei plans to continue working with partners on solutions like this to promote the in-depth integration of AI and industry scenarios and enable the intelligent transformation of ever more industries.

Accelerating Intelligent Transformation White Paper release

Wang also announced the release of a new white paper, Accelerating Intelligent Transformation. It's a collection of case studies and best practices from Huawei, its customers and partners, aimed at helping all industries hit the ground running as they move to embrace new forms of intelligence.

The white paper finds that AI is driving industry upgrade by serving more and more industry scenarios and has become a major engine of growth for social advances. Notably, it also asserts the collaboration between different roles of the business world, academia, and research circles will be important for new AI applications and the development of the entire AI industry. Specifically, these ecosystem stakeholders will need to work together to ensure AI is designed to benefit all by quickly identifying emerging trends, continuously pursuing technological innovation, and rapidly improving both engineering practices. To broaden and deepen AI application across industries, they will need to enable a wide array of models and applications.

The white paper has already received endorsements from several academicians. It reviews both the latest trends and developments in the field of AI, and explores 63 scenario-specific AI applications from 16 different industries. Additionally, it presents a number of innovative best practices in intelligent transformation from 18 different industries.

The theme of HUAWEI CONNECT 2023 is "Accelerate Intelligence". This year's event brings together thought leaders, business leaders, technical experts, partners, developers, and a wide range of industry stakeholders. Together, we will explore how to accelerate industry intelligence through business, industry, and ecosystem development. For more information, visit https://www.huawei.com/en/events/huaweiconnect#Home