MediaTek is hosting the Dimensity Developer Conference, with Generative AI as the main highlight.

On May 7th, MediaTek held the Dimensity Developer Conference 2024 (MDDC 2024) in Shenzhen, with the theme "AI Empowering Everything." Lei Technology was invited to attend and report on the event. Without a doubt, AI was the absolute keyword of this event, and all the sharing and discussions revolved around AI.

In Lei's view, at the Dimensity Developer Conference, MediaTek mainly did three things: launched the "Dimensity AI Pioneer Program" with platform partners and released the "Generative AI Mobile Industry White Paper" with industry ecological partners; brought the edge-side generative AI solution "Dimensity AI Development Kit"; and released the Dimensity 9300+ chip and the key technology Star Speed Engine in the gaming field.

The industry's first white paper has been released, how should we define a generative AI smartphone?

First of all, let's talk about the "Generative AI Smartphone Industry White Paper". In addition to MediaTek, the participants in the formulation also include Counterpoint, Alibaba Cloud Tongyi Qianwen, Baichuan Big Model, Huya, Kugou Music, Zero One Wanwu, OPPO, Soul, Tencent AI Lab, Tencent Hunyuan, and vivo. Among them, there are market research institutions, mobile terminal brands, big models, and internet companies.

The main content of this white paper is to attempt to give a definition to generative AI smartphones and establish a standard. In the white paper, Counterpoint believes that generative AI smartphones need to meet the requirements of supporting local deployment of big models, possessing multimodal capabilities and interaction methods, and providing a natural and smooth user experience.

To meet these requirements, the mobile phone itself needs to have hardware-level computing power that can support the operation of large models, and the chip is undoubtedly the key. MediaTek has always focused on the AI computing power of the Dimensity chip. Currently, both the flagship Dimensity 9300 and the mid-to-high-end market-oriented Dimensity 8200 are equipped with powerful AI computing units APU.

Mobile phones equipped with mid-to-high-end and flagship Dimensity chips can basically meet the requirements of generative phones, as long as the corresponding mobile phone brands can provide more support at the system and application levels, they can bring users a better generative AI experience.

Embracing developers, MediaTek brings the Dimensity AI development kit

In addition to providing SoCs with edge AI capabilities, MediaTek has also introduced the accompanying solution "Dimensity AI development kit." This kit consists of multiple modules, each with corresponding technologies or capabilities.

Among them, the GenAI Model Hub integrates multiple mainstream large models from around the world, becoming a large model library available for developers to choose from; Neuron Studio provides developers with a one-stop visual development environment, greatly reducing development difficulty. The Dimensity AI development kit also supports AI optimization technologies such as inference decoding acceleration and LoRA Fusion.

In simple terms, the Dimensity AI development kit can help developers quickly and efficiently complete the development of edge AI applications, and implement advanced AI technology. Through it, MediaTek can also attract more developers, thereby accelerating the construction of the Dimensity AI ecosystem, allowing more creativity to quickly become applications, and allowing ordinary users to see and touch them.

The Dimensity 9300+ has been released, and the first models have been confirmed.

The Dimensity 9300+ is a natural upgrade from the Dimensity 9300. From a specification perspective, the basic architecture of the Dimensity 9300+ has not changed, and the large core Cortex-X4 has been increased from 3.25GHz to 3.4GHz. One of the highlights of this chip is its AI capabilities. MediaTek has stated that it is the "industry's first" chip to achieve faster Llama 27B edge-side large model operation, with a speed of 22 tokens/second.

It is worth mentioning that the first model to be equipped with the Dimensity 9300 chip is the vivo X100s. We saw this unreleased model in the display area outside the venue. Of course, as it is still under confidentiality, we could only see the appearance through the glass cover and couldn't experience it hands-on. However, it can be confirmed that the AI on the vivo X100s, especially the on-device generative AI, will be the core selling point.

As for the Star Engine, it can help developers better develop games, achieve high performance, high image quality, and low power consumption. Specifically, it includes adaptive technology, which can transmit detailed information about the chip's performance, power consumption, utilization, etc. to developers. Developers can more accurately obtain real-time information about the Dimensity chip, more efficiently call hardware resources, and thus allow games to perform better in terms of performance, power consumption, and other dimensions.

At the same time, the Star Engine also includes ray tracing technology. Just like in PC games, implementing ray tracing in mobile games can bring users a better gaming experience, especially in terms of lighting and shadow effects, resulting in more realistic and delicate image quality, and can also enhance the user's immersion in the game. Of course, when ray tracing is applied, it places higher demands on chip performance and power consumption. The Star Engine can help developers and users in hardware resource balancing and performance scheduling.

The landing of generative AI, applications, and user scenarios are crucial.

Overall, at this developer conference, MediaTek announced a large number of generative AI technology achievements and solutions. However, from the perspective of an ordinary user, Xiaolei is more concerned about the specific application scenarios. In the on-site experience area, Xiaolei experienced some of the Dimensity generative AI technology and specific app integration scenarios.

For example, the Redmi K70E is equipped with the Dimensity 8300-Ultra, which, through the edge-side generative AI capabilities of Dimensity, can achieve AI expansion, image elimination, and other functions. After experiencing it hands-on at the scene, I feel that this feature is still very powerful.

Tianji and Kugou Music have collaborated to achieve fast image matching function. Users can input keywords in the song playing interface to automatically generate an image, meeting the personalized image matching and replacement needs of users.

From the current trend of AI development, it can be seen that both cloud and edge computing have their irreplaceable advantages, and in the future, they will evolve in coordination. As the largest and most user-adhesive smart device, smartphones naturally become a good carrier for edge AI. The strength of edge AI capabilities is directly related to the smartphone SoC. As a leading brand in the mobile chip industry, MediaTek will naturally play a pivotal role in this process.

Through this developer conference, we can feel that MediaTek is not only a chip supplier but also a builder of the AI application ecosystem. It gathers partners from various industries such as mobile phone brands, large model manufacturers, and internet giants, allowing everyone to exert their strengths in their respective fields, thereby driving the development of the entire AI industry. As for the future potential of edge generative AI, it will take more time to provide an answer.