Artificial intelligence is an important technology in the fourth technological revolution. Recently, ChatGPT has been out of the circle, and OpenAI immediately launched a new generation of large language model GPT-4, which once again aroused global attention to the development of artificial intelligence technology. Microsoft announced that it will officially install the GPT-4 model into the Office suite and launch a new AI function, Copliot. In China, Baidu took the lead in launching the generative dialogue product "Wenxinyiyan", and many companies in the finance, media, automobile, medical and other industries announced to become the first batch of ecological partners of Wenxinyiyan. The commercialization of AI is accelerating, opening up more new application scenarios.
█ Reshaping productivity in the era of AI+
AI models can be roughly divided into two categories: decision-making (analytical) AI and generative AI. The decision-making AI model analyzes, judges, and predicts based on existing data. Typical applications are intelligent recommendation content, machine vision, etc. Generative AI, on the other hand, puts more emphasis on deduction and creation after learning and induction, and generates brand-new content. Its essence is to greatly improve and create productivity. The birth of ChatGPT has given birth to AIGC in multiple vertical scenarios, bringing about a huge increase in productivity.
Technology has become more advanced every year, and new directions such as AI large models, generative AI, and general AI have emerged. The adoption rate of AI in enterprises is growing rapidly, and the way enterprises use AI is changing from simple data integration and analytical "+AI" to automated and deep learning "AI+".
Different from the "+AI stage" where AI generates value for traditional applications, the AI+ era emphasizes AI as the core to drive industrial transformation. Enterprises began to try to use generative AI technology to obtain value and new ways of transformation, such as: "AI+Industrial Internet" promotes the transformation of traditional production models to real-time perception, dynamic analysis, scientific decision-making, and precise execution of intelligent production models, becoming an industrial intelligence In the field of service robots, AI technologies such as 3D face recognition, AI visual algorithm, robot learning, navigation technology, and edge intelligence endow robots with stronger autonomous decision-making and learning and reasoning capabilities, creating a new form of human-computer interaction , single-function service robots are gradually being replaced by intelligent service robots with multi-modal and multi-service experiences.
At the macro level, artificial intelligence has become a key field of technological innovation and an important pillar of the digital economy era. In the 14th Five-Year Plan, the country clearly stated that the proportion of the added value of the core industries of the digital economy in GDP should be increased from 7.8% in 2020 to 10%. As the core productivity of the digital economy, artificial intelligence is giving full play to the role of new information infrastructure and effectively stimulating the growth of the digital economy.
█ AIoT applications stimulate end-to-end computing power demand
The birth of AIGC has benefited from the leapfrog progress of the core three elements of computing power, algorithms, and data. In terms of computing power, according to the "White Paper on China's Computing Power Development Index (2022)" issued by the China Academy of Information and Communications Technology, the total scale of global computing power has increased from less than 150 EFlops in 2016 to 615 EFlops in 2021, an increase of more than three times during the period. The growth in intelligent computing power is the most significant.
In terms of data, the IDC report shows that the global data volume has increased from 2ZB in 2010 to nearly 60ZB in 2021, a 30-fold increase in ten years, and is expected to continue to grow to 175ZB in 2025. The huge amount of data provides sufficient support for artificial intelligence. nourishment. In addition, the neural network algorithm based on CNN and DNN has been iterated at a high speed in the past few years, which has greatly shortened the time and cost of algorithm training.
ChatGPT is a dialogue AI model developed based on the GPT3.5 architecture. As a well-known NLP model, GPT is based on Transformer technology. As the model continues to iterate, the number of layers increases, and the demand for computing power increases. Recently, Open AI released the plug-in function of ChatGPT to accelerate ChatGPT's access to third-party applications such as payment platforms, e-commerce platforms, smart home platforms, Internet of Vehicles platforms, and industrial Internet platforms. In the future, AI technology represented by ChatGPT will be integrated with IoT applications, and the demand for computing power will continue to rise. The demand for computing power in the AI era has exceeded Moore's Law.
AIoT is an important direction for the development of the Internet of Things in the future. With the advent of the era of "superman of things", the increase in the number of IoT devices has led to explosive growth of IoT data, and IoT data has become the main force of data generated in modern society. These data with the distinct objectivity of the physical world can provide data sources for AIGC model training, and drive AIGC to learn in-depth knowledge of various industries, thereby generating more accurate information.
Secondly, AIGC can become an important boost for the future development of the Internet of Things towards intelligence. The management of the Internet of Things scene and the interaction between people and things all provide a place for the AIGC large model to come into play. The AIGC model generates customized content according to user needs, creating a personalized experience in IoT scenarios.
AIGC's empowerment of IoT application scenarios will also bring challenges such as network connection and computing power consumption, which also puts forward new requirements for the capabilities of smart modules. On the one hand, there are applications that are very sensitive to delay in the scenario of the Intelligent Connection of Everything, which requires smart modules to provide 5G access to complete on-site millisecond-level low-latency services.
On the other hand, AI technology represented by ChatGPT requires intelligent modules to provide dynamic and efficient algorithms and super computing power support when end-side applications are implemented. Therefore, how to achieve technological breakthroughs in AI computing power, reduce costs, and expand scale will become the focus of smart module technology innovation.
█ MeiG Smart AI Computing Power Module Continues to Evolve
The computing power module is a subset of the intelligent module, developed based on the SoC chip, integrating and fusing a variety of computing units (CPU, GPU, ASIC, FPGA, etc.), including general computing units and high-performance dedicated The computing unit has outstanding advantages in reducing size, reducing cost, reducing power consumption, and being easy to develop.
Compared with intelligent modules with complex systems, the computing power module has sufficient computing power to support the operation of commercial AI algorithms for the computing power requirements in special scenarios, and can configure different communication methods according to different scenarios to achieve better cost performance of products With a higher degree of fit, it can accurately match specific application requirements in different scenarios, and play a powerful role in empowering IoT scenarios related to AI computing capabilities, such as smart cockpit, metaverse, VR, XR, and machine vision. Therefore, "no computing power, no intelligence" has become an important development trend in the era of smart modules 3.0.
The MeiG smart AI computing power module is pre-installed with the Android smart operating system, and integrates core components such as CPU, GPU, NPU, Memory, Transceiver, and PA. It also has rich interfaces and can expand complex peripherals. Customized product development can effectively help customers reduce the cost of the R&D cycle and shorten the time to market. At present, MeiG intelligent AI computing power modules (SNM758, SNM920, SNM930, SNM950, SNM960, SNM970) have gone through three generations of product evolution, forming a matrix product portfolio from 0.2Tops to nearly 48Tops, realizing entry-level, low-to-medium Full coverage of high-end and mid-to-high end.
With the in-depth development of the combination of artificial intelligence and the Internet of Things, MeiG Intelligence keeps pace with the times, and the evolution of its AI computing power modules presents two characteristics: First, the module products are multi-form and can be widely used in smart cockpits, Unmanned aerial vehicle, cloud computing server, sports camera, smart home, smart audio, face payment, video recorder and other fields. Second, the computing power of the modules has been gradually increased to fully meet the high computing power requirements of AI vision algorithms, 3D face recognition, high computing power offline recognition, image semantic segmentation, and robot learning.
Facing all scenarios, MeiG Smart empowers thousands of industries with high-computing AI modules. In the smart factory, under the empowerment of the computing power module, the industrial vision robot quickly completes the collection and analysis of images and data, combines big data analysis and artificial intelligence technology, and optimizes and improves through the analysis and prediction of production data The production process greatly improves production efficiency and quality. For high-definition video surveillance and AI recognition, the computing power module supports multi-channel camera access, and performs intelligent analysis and judgment through visual recognition technologies such as detection, recognition, and tracking; supplemented by AI neural network, it realizes more professional image optimization, and Support 4K or 8K ultra-high-definition video encoding to complete the acquisition of finer picture details. It is of great significance for scenarios such as remote operation, vehicle video surveillance, and smart security.
From technological innovation to product landing is a process of qualitative change to quantitative change. MeiG Intelligent has been focusing on the research and development of high-computing intelligent modules and industrial applications for many years. With profound accumulation, the design and development capabilities in the fields of AI application scenario development, AI performance optimization, and AI low-power program development are at the leading level in the industry, which is a strong technical backing for product upgrade iterations. MeiG Smart has also developed functions such as one-chip multi-screen, multi-camera access, and system virtualization to reduce the difficulty of deploying AI technology and help industry customers quickly apply AI technology.
Big data and big computing power have promoted the development of artificial intelligence to the era of big models. Big models and AIGC have created unlimited possibilities for the digital economy. In the era of AI+, MeiG Smart focuses on technological breakthroughs and application expansion, promotes the continuous evolution of smart modules to higher AI computing power, and builds the computing power cornerstone of AI industrialization, so that AI can achieve inclusive and innovative development.