On May 7, MediaTek held the Dimensity Developer Conference 2024 (MDDC 2024) in Shenzhen, the theme of this MDDC is "AI for all things", and AI is the absolute protagonist of the conference. In his opening speech, Guanzhou Chen, Director, General Manager and COO of MediaTek, shared his views on generative AI.
He believes that generative AI has reshaped the value of end applications and brought new opportunities to the mobile ecosystem. Smart terminals, represented by mobile phones, will become the key carrier for the popularization of generative AI. MediaTek's strength and technology accumulation in the field of edge computing will be able to efficiently empower developers and promote the development of the AI ecosystem.
At the conference, MediaTek and Counterpoint also joined forces with many partners to release the "Generative AI Mobile Phone Industry White Paper", jointly completing the definition of the concept and typical characteristics of generative AI mobile phones. So, will generative AI phones be the key to AI inclusion? Will smartphones gradually converge with AI phones? What role will MediaTek play in this process?
In 2024, the main battlefield of AI will come to the consumer field
In 2023, with the rapid explosion of large model applications represented by ChatGPT, generative AI technology will begin to become well-known to more and more people. Every once in a while, a new batch of large-scale model products emerge on the market, and various generative AI applications such as Wensheng diagrams, Tusheng diagrams, and Wensheng videos are eye-opening for users. In this wave of AI, people are fully aware of the power of generative AI technology.
However, whether it is ChatGPT or Sora, such AI applications mainly focus on productivity and creation. For example, ChatGPT has strong ability to summarize and express words, which is very helpful for writers; Sora's ability to accurately convert text into video is an effective tool for video creators. But frankly speaking, this kind of AI tool is still relatively far away for the more common people. They may experience this kind of AI product out of curiosity and be impressed by the powerful capabilities, but there is not much real user value that can be obtained from it.
In order for emerging AI technology to be popularized and used by more people, it is inevitable to break through narrow fields and target more popular target groups. In other words, AI inclusiveness will inevitably make efforts in the consumer field and take the C-end as the main battlefield. If we say that in 2023, the debut of generative AI will only surprise a few industry users; Then, in 2024, generative AI will be integrated into the daily use scenarios of consumer users and become a technological achievement that the public can directly access.
This is mainly due to the fact that generative AI is an emerging technology that has gradually shown its power, and it has the ability to actively perceive and create content that traditional AI technology does not have. When generative AI technology enters a broader consumer market, it can reach a larger number of users and have a greater impact on their needs and use cases.
In Xiaolei's view, smartphones will become the key to AI inclusion, and the emergence and popularization of AI mobile phones, especially generative AI mobile phones, will allow generative AI technology to enter a huge market of billions of users, and will promote the arrival of the era of AI for all. And why can mobile phones take on such a big role? That's the key topic we'd like to move on.
Mobile phones will be the best vehicle for generative AI
As an emerging cutting-edge AI technology, generative AI can efficiently learn large-scale, high-complexity datasets and complete accurate creations. Generative AI is not only suitable for productivity and creation, but also has huge application potential in more C-end scenarios. Smartphones can become the best carrier of generative AI, mainly because of their unique conditions.
First of all, smartphones are huge. According to IDC data, global smartphone shipments will exceed 1.1 billion units in 2023, making it the world's best-selling smart terminal device. According to a report released by the GSMA, the number of smartphone users worldwide will reach 4.3 billion in 2022, which can be called the most popular smart device.
Secondly, smartphones have extremely high user stickiness. The mobile phone, which combines many functions in one and is highly mature and popular, has become the most frequently used electronic device by the public. Users spend a lot of time on their phones every day, using them for daily communication, gaming, social media, office, and more.
Therefore, if smartphones have generative AI technology and will penetrate into the daily use scenarios of mobile phone users, it will inevitably accelerate the popularization of AI, and mobile phones will be an excellent carrier of generative AI.
At the same time, the smartphone industry, which has entered a period of recession, also needs to introduce new forces to solve its own homogenization problem. After more than ten years of development, the highly mature mobile phone industry is facing the problem of market capacity saturation. The homogenization of smartphone products is more serious, and the replacement cycle of users is also increasing. TechInsights, a research institution, predicts that in 2023, the replacement cycle of global mobile phone users will be extended to 51 months, and the replacement cycle of Chinese users will be relatively short, but there will also be 40 months.
The combination of mobile phones and generative AI will bring more innovation and opportunities to the entire mobile ecosystem. In fact, in the current smartphone market, we can already see some application scenarios for generative AI. For example, many smartphones now add AI retouching functions to their photo albums, such as AI elimination, AI expansion, etc., which improves the efficiency of retouching and enhances the user experience. Some mobile phones integrate large model technology into voice assistants or intelligent assistants, which greatly enhances the capabilities of such tools and brings users a smarter and more personalized interactive experience.
Therefore, mobile phone brands need to embrace AI and implement generative AI to enhance the differentiated competitiveness of the product itself. Ordinary users, after seeing too many homogeneous mobile phone products, are also more willing to choose models with generative AI capabilities and differentiated experiences.
Speaking of which, the key question arises: what conditions should a generative AI phone meet? What are the capabilities?
Since generative AI is a popular technology and can attract users, it will inevitably be used by mobile phone manufacturers as a marketing selling point. Once this phenomenon is rampant, there will inevitably be a mix of real and fake AI mobile phones in the market. In the long run, when bad money drives out good money, the market will lose confidence in generative AI technology and products, which is not conducive to the continued popularity of AI.
At this time, it is particularly important to formulate specifications for generative AI mobile phones and influence the behavior of market players. The value of the "Generative AI Mobile Phone Industry White Paper" we mentioned at the beginning is reflected at this time. In Xiao Lei's view, this white paper has two more significant characteristics, one is authoritative enough, and the other is comprehensive enough.
The members who participated in the formulation of the "Generative AI Mobile Phone Industry White Paper" include chip manufacturer MediaTek, market research agency Counterpoint, Alibaba Cloud Tongyi Qianwen, Baichuan Model, Huya, Kugou Music, Zero One Everything, OPPO, Soul, Tencent AI Lab, Tencent Mixed Yuan, and vivo, all of which are leaders among mobile phone manufacturers, large model brands, and Internet manufacturers, and are deeply involved in the generative AI industry. Therefore, its authority is sufficient.
In addition, this white paper defines the concept of generative AI mobile phones, explains the AI strategies of chip design companies, mobile phone OEMs, and developers, analyzes and summarizes the current status of generative AI mobile phones, and predicts the future of related industries. Such a white paper that clearly introduces the technical concept, market status and industry strategy can help users and partners learn more about generative AI and generative AI mobile phones.
The content of the "Generative AI Mobile Phone Industry White Paper" has actually answered the question just raised, and it points out that the key to generative AI mobile phones is to have basic capabilities such as deploying large models on the device side, being able to interact in a natural way, and supporting multimodality. According to Xiaolei's understanding, a genuine generative AI mobile phone needs to have local computing power in the hardware to support the operation of large models on the device side, provide support for new AI interactions at the system level, and realize the implementation of multimodal application scenarios.
Among them, the most basic is that the mobile phone needs to have the hardware strength of the device side AI deployment, in other words, the mobile phone needs to have strong enough AI computing power. Therefore, generative AI mobile phones will inevitably put forward higher requirements for the AI computing power of SoCs. And this MDDC, an unexpected surprise brought by MediaTek is the release of the flagship chip Dimensity 9300+. The chip's CPU and GPU performance are strong enough, but most importantly, it has powerful generative AI capabilities. In addition to its powerful computing power, the Dimensity 9300+ is also the first to support AI speculative decoding acceleration technology and Dimensity AI LoRA Fusion 2.0 technology on the device side, which can provide a more efficient and personalized generative AI experience. It supports mainstream generative AI models and can provide end-to-end generative AI multimodal applications such as text, images, and music.
It is no exaggeration to say that flagship chips such as Dimensity 9300+ not only lay the foundation and build the framework for the house of generative AI mobile phones, but also provide decoration solutions, which greatly reduces the difficulty of landing generative AI mobile phones. When MediaTek joins forces with mobile phone brands and developers, it can create generative AI phones with strong AI power and excellent innovative experience.
MediaTek will be the leader in generative AI mobile phones
In fact, MediaTek's layout in the field of mobile phone AI is much earlier than many people think. Back at MWC in 2018, MediaTek released the Helio P60. The mobile phone chip is positioned in the mid-range, but it is equipped with an independent AI computing unit APU, and it debuts MediaTek's NeuroPilot AI technology, which can use the capabilities of CPU, GPU and APU for AI computing. At that time, MediaTek's mobile AI technology was mainly used in the field of imaging. Although there were many differences between AI technology at that time and generative AI at present, the advance layout of AI still laid the foundation for MediaTek's subsequent generative AI technology explosion.
In 2021, with the debut of the Dimensity 9000, MediaTek officially sounded the clarion call to enter the flagship mobile phone chip market. The core highlights of the Dimensity 9000 are not only comprehensive enhancements to the process, CPU, and GPU, but also energy-efficient APUs. It is equipped with an APU 590, which adopts an energy-efficient AI architecture, which greatly improves both AI computing power and energy efficiency. The Dimensity 9000's AI computing power has been successfully applied to mobile phone usage scenarios such as video, gaming, and video, bringing users a better experience.
In 2022, the Dimensity 9200 was released, which further enhanced AI performance, and the integrated APU 690 adopted a hybrid computing + intelligent neural network architecture, improving AI performance by 35%. The Dimensity 9200 supports intelligent image semantic technology, which uses artificial intelligence to analyze environmental colors, object construction, and motion, layer labels, and adjust colors to improve overall image quality. Based on the in-depth integration of APU and ISP, the Dimensity 9200 also derives image-related functions such as AI dual-track capture.
The Dimensity 9300 released in 2023 and the Dimensity 9300+ unveiled at the MDDC conference have comprehensive generative AI capabilities, and the support for device-side AI and large model access has been unprecedentedly enhanced. It is not difficult to find that in terms of generative AI technology, the strength of Dimensity chips today is not made in a day, and it is inseparable from MediaTek's advance layout and continuous investment, MediaTek's deep accumulation in the field of edge computing and mobile phone chip research and development allows it to seize the trend of generative AI in time and become the leader of generative AI chips and generative AI mobile phones.
In addition, data provided by Counterpoint shows that in Q4 2023, MediaTek won 36% of the global mobile phone chip market share, ranking first. MediaTek has won the global mobile phone chip shipment championship for 14 consecutive quarters, which is enough to see MediaTek's leading position in the mobile phone chip industry. It can be seen that at present and in the future, a considerable number of models in the mobile phone market are equipped with Dimensity chips, and MediaTek's trends in generative AI technology will directly affect the entire generative AI mobile phone industry.
It is worth noting that MediaTek is not only a provider of mobile phone chips and AI computing power for generative AI mobile phones, but also a leader in the construction of a generative AI ecosystem and provides all-round support for developers. At this MDDC, MediaTek launched the Dimensity AI Development Kit, which provides a full range of solutions for end-to-end AI development, helping developers efficiently implement their ideas, greatly reducing development costs, and quickly delivering AI applications to users.
In MediaTek's view, whether it is AI chips or generative AI technology itself, it is actually difficult to directly act on C-end users. Just as the development and popularization of smartphones in the past, the key to the development and growth of generative AI mobile phones today lies in the ecosystem. MediaTek launched chips with all-round generative AI capabilities, created end-to-end AI solutions for developers, launched the "Dimensity AI Pioneer Program" with partners, and released the "Generative AI Mobile Phone Industry White Paper", all with the core purpose of creating a good environment for the generative AI ecosystem and cultivating a soil for the seeds of innovation to grow and sprout.
Why can MediaTek become a leader in generative AI mobile phones? Xiao Lei believes that this mainly comes from its strength and insight.
First of all, it has strong hard power, and MediaTek, as a chip manufacturer, has the ability to create chips with strong computing power and comprehensive support for AI scenarios to meet the end-side computing power required by the sweeping wave of generative AI. The Dimensity chip, represented by the Dimensity 9300+, will become a solid hardware base for generative AI mobile phones. MediaTek's past technology accumulation in the field of AI and its continuous cultivation in the market have now become valuable technology accumulation and confidence. After years of continuous iteration and updating, the APU on today's chip has achieved both high computing power and high energy efficiency, maintaining a leading position in the mobile phone chip market.
The second is keen insight, including a keen insight into the market and the needs of developers. MediaTek has launched a more developer-friendly Dimensity AI development kit, which includes GenAI best practices, GenAI Model Hub, GenAI optimization technology, and Neuron Studio.
GenAI best practices provide developers with advanced technologies in terms of model quantization, compilation, and inference, and can also open the source code for device-side deployment, accelerating the terminal deployment speed of large models from weeks to one day. GenAI Model Hub provides developers with global mainstream models, including Meta Llama, Baichuan Model, and Ali Tongyi Qianwen, providing developers with a wealth of model choices.
At the same time, the powerful GenAI optimization technology introduces the industry's latest and fastest speculative decoding acceleration and software optimization technologies such as LoRA Fusion, which significantly improves the generation speed of generative AI applications and reduces the storage space used. In addition, the Neuron Studio brought by the Dimensity AI Development Kit organically integrates MediaTek's various advanced tools in generative AI and integrates a complete development environment, providing developers with a one-stop, visualized and easy-to-use development environment.
MediaTek understands the problems that developers encounter in generative AI application development and the capabilities they need to rely on, so that it can develop a suite around the needs of developers and provide a "fast, complete, strong, and easy" professional development experience. This also allows the industry to see MediaTek's determination and sincerity to accelerate the construction of a full-scenario AI application ecosystem.
Moreover, at this MDDC, the generative AI applications demonstrated by MediaTek and its partners are closely related to the daily use needs of ordinary users.
For example, the AI expansion function on the Redmi k70E, the generated image is very natural to connect with the original picture, it is difficult to see that it is generated by AI, and the effect of being fake and real is achieved; Using the application of LoRA fusion technology, the captured real portraits can be quickly converted into cartoon-style characters, bringing users a personalized experience. Only when generative AI is always connected to the actual needs of users can this technology take the road more stable and wider.
More importantly, while the mobile industry is still discussing how to better deploy large models to smartphones and other smart terminals, MediaTek has been working with industry partners to create a full-scenario generative AI application ecosystem, which also reflects MediaTek's fundamental purpose of starting from user experience.
At this conference, MediaTek united various partners in the AI industry, including mobile phone brands, large model brands, Internet manufacturers, etc., to gather the strength of all parties to jointly promote the development of the entire industry. Judging from the presentation of the conference and the display content in the exhibition area, there have been many application scenarios using Dimensity generative AI technology and development solutions. In terms of user reach and perception, MediaTek has done a very good job.
Of course, Rome wasn't built in a day, and the big garden of mobile generative AI ecosystems won't prosper in a short period of time. However, as a provider of top-notch end-side AI computing power and a leading builder of the generative AI ecosystem, MediaTek has cultivated a soil suitable for generative AI innovation, providing developers with the conditions to unleash their creativity, and the generative AI ecosystem is developing and improving in the right direction. It is foreseeable that in the future, MediaTek will continue to iteratively upgrade its generative AI capabilities on Dimensity chips, work with partners to cultivate the generative AI ecosystem, accelerate the popularization of generative AI mobile phones and related application scenarios, and continue to lead the development of the entire industry.