If 2023 is the first year of generative AI, then 2024 is the year of accelerated adoption of generative AI.
After the release of ChatGPT, artificial intelligence has entered the era of large models. The advent of generative AI is beginning to propel humanity into the stage of digital civilization.
As Guo Wei, Chairman of Digital China, said: "In the next ten years, the corporate strategy of all enterprises will make full use of the three natives (cloud native, digital native, and AI native) to subvert their own business, and Digital China is also willing to become a partner in the full life cycle of digital transformation of enterprises, let us meet such a great era together." ”
01
Information innovation and AI computing power build a foundation for intelligent computing
Information innovation, that is, the information technology application innovation industry, is the foundation of data security and network security, an important part of new infrastructure, and the "engine" of digital upgrading in all walks of life. Based on information innovation, Digital China continues to build its own brand of "Shenzhou Kuntai", forming three product lines covering "data computing, terminal products, and data networks", and continues to deploy in "localized computing power + intelligent computing".
In the past year, a series of self-owned brand products such as the Shenzhou Kuntai all-in-one machine and the KunTaiA924 all-in-one AI server have been launched. 14 Shenzhou Kuntai servers passed the "Special Evaluation of General Server Government Procurement Demand Standards".
With the development of AI, the amount of computing data and the scale of parameters used by large models have increased exponentially, which has led to an explosive increase in the demand for intelligent computing power. IDC predicts that the scale of intelligent computing power in China is expected to enter the level of 10 trillion floating point calculations per second in 2026, and the compound annual growth rate of intelligent computing power is expected to reach 52.3% from 2021 to 2026.
Han Zhimin, Vice President of Digital China and Chairman of Digital China Information Innovation Holdings
Han Zhimin, Vice President of Digital China and Chairman of Digital China Information Innovation Holdings, also said: "At present, the development of AI is facing huge computing power challenges. "And with that, a range of infrastructure-level problems and challenges have emerged.
On the one hand, there are complex compatibility and utilization issues between and within intelligent computing clusters. At present, domestic enterprises are in a complex business environment, and heterogeneous intelligent computing infrastructure has become inevitable, but there is still a lot of room for improvement in the model floating-point utilization rate that can be achieved by heterogeneous intelligent computing clusters. For example, when OpenAI trained ChatGPT4, the usage rate of MFUs was only between 32%~36%. At present, the average industry utilization rate of FFU utilization rate is 30%~40%, and it is extremely difficult to increase it to 50%. As a result, China Kuntai has launched the HISO heterogeneous intelligent computing scheduling and operation platform and the new heterogeneous intelligent computing convergence acceleration platform HICA.
Based on 100% cloud-native technology, the Heterogeneous Intelligent Computing Scheduling and Operation Platform (HISO) integrates GPU hard sharding and virtual sharding technologies to realize the pooling and cross-cluster scheduling capabilities of GPU resources, and improve the resource utilization of GPU server clusters.
The Heterogeneous Intelligent Computing Convergence Acceleration Platform (HICA) has the characteristics of one cloud and multiple cores, supports mainstream AI chips at home and abroad, such as Huawei's Ascend, NVIDIA, and Intel, and can realize hybrid training and inference of training and inference tasks in intelligent computing clusters composed of different brands and models of chips, and is expected to reduce the idle computing power of GPUs by 20%.
On the other hand, generative AI models have a large number of parameters, and computing power and high energy consumption have become the core problems that need to be solved. In order to meet the computing power requirements in the intelligent computing center, China Kuntai has built a variety of AI servers, covering all scenarios from training to inference.
神州鲲泰全液冷整机柜方案 Kuntai post 2000
In response to the pain points of high energy consumption in the intelligent computing center, Shenzhou Kuntai has also launched an all-liquid-cooled cabinet delivered in an integrated manner, which does not require installation and commissioning on site, and the efficiency is increased by 10 times. The maximum single cabinet power is 60kW+, and the traditional 4~8 cabinets are at the top of one cabinet, which is 1.5 times the energy efficiency ratio of domestic competitors.
In the past year, the construction of the Intelligent Computing Center has been in full swing in China.
At the policy level, "new quality productivity" was written into the government work report for the first time, the "artificial intelligence +" action was fully launched, and the SASAC held a special meeting on "AI empowering industry renewal", which proposed: accelerate the construction of a number of intelligent computing centers.
In 2023, there will be more than 100 new intelligent computing center projects across the country, with a budget of hundreds of millions. Since the launch of the "Eastern Data and Western Computing" project, traditional IT companies, cloud vendors and telecom operators have been intensively deploying intelligent computing centers.
As one of the important participants in the "Kunpeng + Ascend" computing industry ecosystem, Digital China is deeply involved in the Intelligent Computing Center, and provides professional infrastructure hardware for the Intelligent Computing Center. At present, Digital China and Digital China Holdings have participated in the construction and operation of Changchun New Area Intelligent Computing Center, Shenyang Artificial Intelligence Computing Center, Xiamen Kunpeng Supercomputing Center, and Hong Kong SAR Government Large Model Intelligent Computing Center.
It is worth mentioning that in January this year, in order to provide the most basic technical environment for ecological partners and focus on AI computing power, the Digital China Shenzhen Artificial Intelligence Computing Center project has been officially launched. It is expected that in the near future, this intelligent computing center equipped with the Shenzhou Wenxue platform will officially provide computing power, models, applications, and solutions related services to the market and the ecosystem.
Judging from the financial report, the entry of AI and the joint force of products have enabled Digital China's information innovation business to achieve explosive growth. In 2023, the overall revenue of Digital China's information innovation business will exceed 3.4 billion, maintaining high growth for three consecutive years. In the first quarter of 2024, the single-quarter revenue scale of Kuntai's AI computing power-related business is close to that of last year.
02
AI native scene empowerment, "to the vast and subtle"
In the past year, AI large model technology has developed rapidly, and how to make good use of large models to reduce costs and increase efficiency and promote business growth has become a real concern for enterprises.
In this era of "100 model wars" or even "1000 model wars", how to transform technology into real productivity? The answer is to break the situation with the power of ecology and accelerate the implementation of generative AI at the enterprise level.
In 2023, the launch of Shenzhou Wenxue will become a new paradigm to help enterprises implement AI.
At present, MDA has connected with dozens of mainstream large models, and released agile applications such as smart reading, smart analytics, and smart Q&A, successfully helping customers in pharmaceutical, retail and other industries to implement generative AI application scenarios.
At the Digital Cloud Force Conference 2024, a new version of Shenzhou Wenxue was released, including three modules: Agent Engineering, Enterprise Knowledge Governance, and Model Training and Management.
Taking the three functional sections of the new version of Digital China as the starting point, Digital China has its own considerations: in reality, enterprises have a large number of scenarios that need to be empowered by AI. For enterprises, how to better integrate AI scenarios with each business, and how to make internal IT departments and business departments cooperate more intelligently, cannot be met by a simple developer platform alone. Therefore, enterprises need efficient and complete AI application development frameworks and specifications that have been proven in practice. Requires continuous precipitation of tools; The continuous deepening of knowledge governance requires the continuous training and optimization of several enterprise models. At the same time, in order to ensure data security, this needs to be done in the private domain of the enterprise. From this point of view, the AI-native innovation of enterprises needs a platform to continuously empower and accelerate. Shenzhou Wenxue is such an AI-native application empowerment platform, and all three major sectors cover the needs of enterprise customers.
神州数码副总裁、CTO李刚
Li Gang, Vice President and CTO of Digital China, said: "When 'To the Masses' has been fully expressed, Digital China has become more and more determined to focus on the customer-centric 'subtlety', which is the top priority to truly realize the landing of AI. ”
Regarding the enterprise data security that everyone is concerned about, Li Gang shared in an interview that Shenzhou Ask learns to protect enterprise data security through three aspects.
First, the scenario of Shenzhou's deployment is a privatization scenario. Deploy in a fully customer-controlled environment. Second, the large models managed by Shenzhou Wenxue, whether commercial or open-source, are also localized after fine-tuning and pre-training. In these two points, the basic security issues can be guaranteed. The third is how to ensure the accessibility of knowledge within an enterprise. Enterprises will need multiple models due to different departments, such as human models, financial models, etc., and the key is to give enterprise data a "security fence". "Make sure the model answers the right questions that it can answer and doesn't answer irrelevant questions," Li said. "Adding this together will further ensure the security of enterprise data.
Shenzhou Kuntai all-in-one machine
In addition, we also see that Digital China has given full play to its independent innovation capabilities in "computing power + AI" software and hardware products, and has built a Shenzhou Kuntai all-in-one machine. Based on the KunTaiA924 integrated AI server for training and pushing, training and inference can be seamlessly connected and carried out at the same time, greatly improving the training efficiency of large models. Pre-installed Digital China's self-developed generative AI product "Shenzhou Wenxue", which is ready to use at start-up, simplified deployment, no tuning, and rapid deployment of enterprise-specific AI models.
03
The AI industry is accelerating, and ecological innovation is being aggregated
The advent of Chat GPT has brought generative AI to everyone's eyes overnight.
Every business is anxious about this revolutionary technology, hoping to find its own opportunities to innovate. But in the process of exploring AI, there are still many practical problems ahead.
On the demand side, many companies are at a loss for the speed of iteration of generative AI, and they need to find a balance between investment and return. On the supply side, in the process of commercialization, it is sometimes difficult for a single enterprise to combine the entire AI technology stack with the industry needs of customers. In terms of talent, the shortage of talent is the biggest obstacle to the continuous development of artificial intelligence.
Wang Bingfeng, Co-Chairman and CEO of Digital China
Wang Bingfeng, Co-Chairman and CEO of Digital China, said: "With these real-world issues and the huge disruptive potential of artificial intelligence that we see in the future, no company can move forward alone. We believe that the ecosystem will become an indispensable and powerful driving force for the acceleration of AI implementation. ”
Digital China hopes to build a new type of ecological partnership, starting from three aspects: one is the ecosystem of solutions and technologies, the second is the ecology of talents, and the third is the ecology of the market.
In terms of solutions and technology ecology, technology itself will not disrupt anything, only solutions, applications and business models that can be implemented will. Native AI applications can solve problems that could not be solved in the past. Only when everyone works together to produce enough AI-native solutions and applications can a healthy ecological environment be formed. Evolve AI from an enterprise tool to an independent agent for enterprises.
In terms of talent ecology, Digital China will work with the most leading artificial intelligence companies at home and abroad to jointly build an artificial intelligence training and certification system. At present, Digital China is the only training partner in Chinese mainland that has three officially authorized training partners of Microsoft, AWS and Google, as well as training partners of dozens of advanced enterprises such as Alibaba Cloud, Tencent, Tianyi, Huawei, Huawei Cloud, Oricle, and IBM.
In terms of market ecology, relying on Digital China's 2035 Lab and Artificial Intelligence Research Institute, Digital China will continue to strengthen cooperation with universities, institutes, institutions and associations to jointly promote the long-term development of artificial intelligence in the technology ecology.
04
epilogue
Generative AI has been on the run for nearly a year, and it's still on the rise.
The clarion call for the AGI era has been sounded, and Digital China will focus on three things: AI native scenario empowerment, multi-cloud heterogeneous green intelligent computing, and international AI ecological innovation.
At the end of the opening ceremony of the Digital Cloud Force Conference, Guo Wei, Chairman of Digital China, said: "We put forward a new proposition for Digital China in the AGI era: a customer-centric AI landing acceleration engine. Data-cloud integration has also reached a new level, upgraded to AI-driven data-cloud integration. Digital China will continue to explore the path of "artificial intelligence+" in China's modernization, and carry out more interactive and collaborative innovation with the international innovation center, a new landmark of AI in the future. ”