Build the foundation of the industry and build a digital intelligence ecosystem

[Shenzhen, China, November 9, 2022] Huawei Connect Conference 2022 will be held simultaneously offline and online in Shenzhen. Huawei and its partners jointly build an industrial foundation to help the rapid development of the digital intelligence ecosystem. From “available” to “easy to use”, Kunpeng fully serves the core scenarios of the national economy and the people’s livelihood industry. Ascend AI’s basic software and hardware platform has been fully upgraded to help scientific computing enter a new stage of scientific intelligence. Huawei proposes to establish a scientific and intelligent innovation consortium to join hands with the industry to build a leading pattern of scientific intelligence.

Deng Taihua, Vice President of Huawei and President of the Computing Product Line, delivered a speech

Deng Taihua, Vice President of Huawei and President of the Computing Product Line, delivered a keynote speech on “Building the Industrial Foundation and Building a Digital Intelligence Ecosystem Together” . Wisdom foundation, build an industrial ecology together, and create an era of digital intelligence.

Kunpeng and Shengteng, together with Hongmeng, HMS, HUAWEI CLOUD, and MDC for intelligent driving, empower partners, and the digital intelligence ecosystem has developed rapidly: more than 320 million Huawei terminal devices equipped with HarmonyOS; HMS overseas Top 3000 applications integration 80%; the total number of HUAWEI CLOUD developers exceeded 3.5 million; the number of Kunpeng and Ascend partners exceeded 5,000, the number of solutions reached 13,000, and the total number of developers exceeded 2.6 million.

Digital foundation, Kunpeng breaks through the depth of the industry

Kunpeng’s industrial ecology has developed steadily, with hardware partners accounting for more than 95% of shipments. Euler has realized the transition from enterprise leadership to industrial co-construction, and quickly crossed the ecological inflection point. The current cumulative installed capacity has reached 2.45 million sets. First. Kunpeng development kit DevKit and application enablement kit BoostKit are continuously optimized to improve native development capabilities and help developers develop minimally. By the end of this year, the ecological satisfaction degree in the national economy and people’s livelihood industry will exceed 70%. Currently, more than 4,200 partners have been developed, and 12,000 projects have been released. A number of Kunpeng certification solutions have been applied on a large scale in industries such as government, finance, transportation, and electric power.

The foundation of intelligence, Ascension makes breakthroughs in industrial development

Huawei focuses on innovation in the foundation of intelligence, and continues to consolidate the Ascend AI basic software and hardware platform around “extreme performance and ease of use”. CANN 6.0 opens the underlying development system, and the efficiency of custom operator development is doubled. Based on the continuous enhancement of large model capabilities, Ascension MindSpore 2.0 has been upgraded to an AI fusion framework that natively supports scientific intelligence. #Ascension AI Upward Power#

Based on the basic software and hardware platform of Ascend AI, Ascend training and inference solutions have been newly upgraded, and the ease of use and performance in training and inference scenarios have been comprehensively improved. In the training scenario, the typical model development cycle is less than 0.5 person-week, and its performance is 20% ahead of the industry. The reasoning solution supports one-stop full-process tooling, which doubles the development efficiency. After the model is deployed, it can be automatically incrementally trained, with continuous iteration, without business interruption. Accuracy keeps improving.

Ascend’s AI capabilities have been comprehensively improved to facilitate the construction of artificial intelligence computing centers. At present, more than 20 computing power nodes have been connected to China’s computing power network, including artificial intelligence computing centers, super computing centers, etc., with a total computing power of more than 3000P. Ecology is shared across the entire network.

Based on large computing power, the AI ​​large model full-process enabling system supports large model innovation from planning, development, and industrialization. In the development stage, based on Ascend’s large model development enabling platform, large models are easier to develop, adapt and deploy; in the industrialization stage, the large model industry will be solved through the industrial consortium of remote sensing, multi-modality and fluid mechanics. Application landing and industrial incubation will eventually serve thousands of industries.

China AI large model map pre-release

The current trend of artificial intelligence technology is developing in the direction of general-purpose large models. Large models have stronger generalization capabilities and can cover multiple business scenarios. The development of large models has also become a consensus of industry, academia and research. Huawei’s AI large model full-process enabling system supports large model innovation from planning, development, and industrialization. Under the guidance of the Ministry of Science and Technology, the China Institute of Science and Technology Information cooperates with the industry to jointly plan a large model map, and will soon release the “China Artificial Intelligence Large Model Map Research Report” to coordinate the orderly development of China’s large model, and explore in the form of industrial consortia. New methods and new models for the implementation of large-scale models, jointly create benchmark projects, and accelerate the promotion of large-scale models to effectively enable the development of the industry.

With the introduction of AI, scientific computing has entered a new stage of scientific intelligence

Scientific intelligence involves two major computing fields, HPC and AI, covering three major computing scenarios: AI-enabled mechanism computing, data-driven AI computing, and the integration of mechanism computing and AI computing. In the past, a single, chimney-shaped software and hardware platform could no longer meet the needs of scientific intelligence. Based on Kunpeng and Ascend AI, Huawei has built a native scientific intelligence basic software and hardware platform through an innovative converged computing architecture to accelerate application innovation. At present, it has been applied in new material research and development, large aircraft design, protein structure prediction and other fields.

In order to better focus on industrial value scenarios and open up channels from scientific research innovation, application demonstration to industrial promotion, Huawei proposes to establish a scientific and intelligent innovation consortium to join hands with industry, academia and research to break through the frontiers of science, accelerate the process of scientific research and innovation to industrial implementation, and strengthen interdisciplinarity. Construction and personnel training, and jointly build a new ecology of science and intelligence.

Talent foundation helps digital intelligence ecology flourish

Since the launch of the “Smart Base” production-education integration collaborative education base in 2020, Huawei, together with the Ministry of Education and 72 colleges and universities, has cultivated more than 5,000 pioneer teachers, opened more than 1,500 courses, and participated in a total of 500,000 students learning and practice.

Facing the future, Huawei will continue to provide course resources, teacher training, and technical expert support, develop more basic technology textbooks, train more pioneer teachers, and expand more teaching methods, from Kunpeng, Ascend, Huawei Cloud, to 5G, optical The entire ICT field, such as ICT, data communication, and storage, cultivates a steady stream of innovative talents for the development of the digital intelligence ecosystem.