In the past year, the most popular keyword in the field of consumer electronics is terminal Side AI.

Large models have powerful understanding and generation capabilities; terminal devices are the first choice for interaction between humans and AI. The collision of the two makes Duan Side the best way to land a large model, if not the only one. More and more consumer electronics terminal manufacturers such as mobile phones, PCs, and automobiles are applying AI technology to their products and services, and the prosperity of end-side AI is inseparable from a core support point—the computing platform (chip).

End-side computing power comes from the core. Therefore, high-performance chips are the source of the continued prosperity of end-side AI, and it is also the direction Qualcomm hopes to promote through technological innovation and ecological cooperation.

The Snapdragon Summit was held not long ago, we see Qualcomm continues to strengthen its terminal-side AI product layout and support for domestic large models. It not only announced cooperation with large model manufacturers such as Zhipu and Tencent Hunyuan to jointly promote the deployment and implementation of end-side AI models. In addition, it also launched the Snapdragon 8 Extreme Edition mobile platform for AI phones and the Snapdragon Extreme Edition automotive platform for smart cars.

An Meng, President and CEO of Qualcomm, said that AI, especially generative AI, is one of the biggest changes in the field of mobile computing; Qualcomm focuses on generative AI, especially how AI changes the computing experience, edge side How it will develop and how it will bring about a disruptive innovation cycle, we are committed to bringing AI-first experiences into every computing space.

The trend of the era of client-side AI originates from mobile computing represented by Qualcomm, which deeply embraces AI. We might as well start with Qualcomm’s direction of creating an AI-first experience to find the source of power for device-side AI to take a step forward.

At the beginning of 2024, consumption of PCs, mobile phones, cars, etc. Electronic categories have begun to actively combine with large models to create new AI XX hardware, and device-side AI has entered a stage of concentrated explosion.

It can be said that large models have opened up a new trend in the field of consumer electronics, and if we go deep into the center of the trend, we will see a key force that promotes the brewing and maturity of end-side AI - mobile computing.

The irreplaceability of on-device AI to cloud-side AI is that large models are calculated locally, avoiding data uploading to the cloud, and solving concerns about user data and privacy security; it is more real-time and suitable for games and photography. , AIGC and other applications have lower processing latency on the client side, faster feedback, and better experience.

Entering 2024, OpenAI, Zhipu, Tencent Hunyuan and other first-tier teamsMany model manufacturers have realized the value and constraints of large-scale models on the ground, and have launched smaller-scale, higher-performance model versions one after another. Although large models are actively being miniaturized, due to limitations such as end-side computing power and memory space, model reduction and compression will also compromise the performance of standard general-purpose basic large models.

The device-side deployment of large models relies heavily on the device-side computing platform to provide powerful performance, support ultra-large-scale data operations, and achieve a balance between performance and power consumption.

Allowing large models to better adapt to the device side has become a top priority in the field of mobile computing, and Qualcomm is actively solving it.

At the recent Snapdragon Summit, Qualcomm announced its cooperation with Zhipu and Tencent Hunyuan to jointly promote the deployment and implementation of device-side AI models. We can see three changes in mobile computing + large models from Qualcomm’s new actions:

Change 1: Richer device-side AI capabilities. The NPU performance of the Snapdragon 8 Extreme Edition is 45% higher than that of the previous generation. The powerful performance can run multiple AI workloads at the same time, which means that more AI applications based on large models can be deployed on the device side.

Change 2: Better client-side smart experience. Qualcomm equips the Snapdragon 8 Extreme Edition with the highest-speed LPDDR5x memory currently available, which solves memory concerns. Currently, Tencent Hunyuan is cooperating with Qualcomm to promote the deployment of 7B and 3B Hunyuan large models in terminals.

Change 3: A more robust device-side AI ecosystem. AI developers are an indispensable role for the continued prosperity of device-side AI. Qualcomm has prepared efficient development tools based on its own hardware for developers, including Qualcomm AI software stack (AI Stack) and AI Hub, which work closely with Tencent Hunyuan and Zhipu AI. Reaching cooperation will help model manufacturers better release platform computing power on Qualcomm hardware and support AI developers to easily integrate AI models into applications.

According to the data released by Qualcomm at the Snapdragon Summit, making full use of the powerful terminal-side AI performance of Snapdragon 8 Extreme Edition and the performance optimization brought by Qualcomm AI software stack to the model, Zhipu GLM-4V terminal-side vision Large models can run at high speed on the terminal side at over 70 tokens/second. Optimizing the client-side performance of large models will not only reduce developers’ usage costs and improve their product effectiveness, but will also contribute to the commercialization of basic general-purpose large models.

Large models are the starting point of the era of client-side AI; mobile computing provides power guarantee for large models on the client side in terms of more applications, better experience, and stronger ecology.

Qualcomm’s strong support for large models and the industry prosperity of end-side AI from the source are one of the manifestations of promoting AI-first experience.

Large models must be used by users on terminal devices in order to A technological trend that transforms into the industrial prosperity and commercial value of end-side AI. When starting a large AI model, hardware performance is the foundation.

We interviewed the person in charge of an AI mobile phone brand who said: AI has just begun in the mobile phone industry. In the next year, the AI ​​experience will exceed your imagination, and the device-side model has very high performance requirements. , so performance is never excessive.

Where does performance come from? The answer is computing platforms. Regarding the two major categories where end-side AI is currently concentrated: one is personal mobile terminals, such as AI mobile phones and AI PCs; the other is intelligent connected cars, where AI applications such as smart cockpits, assisted driving, and autonomous driving are all accelerating. Qualcomm brought the latest technologies and products of the two major platforms at the recent Snapdragon Summit.

Mobile platform:

The Snapdragon 8 Extreme Edition chip has brought a major leap in the intelligence of AI mobile phones with its technological performance breakthroughs and innovative capabilities.

Snapdragon 8 Extreme Edition generates AI on the terminal side Breakthroughs in , multi-modal processing and image processing can provide support for a large number of AI use cases in AI mobile phones. For example, the revolutionary breakthrough of AI-ISP technology, closely integrated with NPU, makes real-time processing of high-resolution image data possible. Mobile phone manufacturers can introduce more AI applications into photography functions, such as automatic scene recognition, real-time beautification and automatic Focus, etc., this is one of the product capabilities that young consumer groups value particularly today.

In addition, the Snapdragon platform also actively supports AI applications in the PC field.

At the just-held 2024 World Internet Conference, Qualcomm won the "World Internet Conference Leading Technology Award" with its Snapdragon X Elite platform designed specifically for Windows 11 AI+ PCs. One of the core advantages of Snapdragon X Elite is the use of customized Oryon CPU, which has extremely fast response speed and can provide strong performance for AI PC. For example, Honor's latest AI PC, with the support of the computing power of Snapdragon X Elite, realizes AI functions such as AI minutes and AI two-way noise reduction, which greatly improves users' meeting and communication efficiency. Currently, the Snapdragon X series supports 58 PC products that have been released or are under development, strongly supporting the growth of the new category of AI PC.

In terms of automotive platforms:

Automotive intelligence has become a general trend, among which smart cockpits and smart driving have become the core competition points of car companies. To provide the underlying power for smart cars, Qualcomm has launched the new Snapdragon Cockpit Extreme Edition platform and Snapdragon Ride supreme version platform.

Among them, the Snapdragon Cockpit Extreme Edition platform also uses Qualcomm's self-developed Oryon CPU, specially designed for automobiles, has greatly improved performance and can help car companies integrate capabilities such as large models and AIGC with the cockpit to create new experiences such as AI voice assistants, AIGC applications, complex situation awareness, and natural language interaction.

The Snapdragon Ride Extreme Edition platform is also based on Qualcomm's most advanced NPU and Oryon CPU, bringing better intelligent driving to cars, achieving high-precision positioning, environmental perception, decision-making and planning, and vehicle control. series of intelligent driving functions.

It is also worth mentioning that to enhance the product capabilities of AI mobile phones, AI PCs, and smart cars, more powerful mobile computing hardware is only a necessary condition. In addition, manufacturers and developers also need a series of development work to release hardware performance and optimize device-side AI performance.

Currently, the Snapdragon hardware platforms used in mobile phones, PCs, automobiles, etc. have integrated Qualcomm’s self-developed Oryon CPU. Coupled with the empowerment of Qualcomm AI software stack, it is expected to realize efficient terminal-side AI applications on various platforms. development.

Hardware (mobile platform, PC platform and automotive platform) and software (AI Stack, AI Hub) collaborate, Qualcomm is removing the performance constraints of end-side AI, driving various consumer electronics products such as mobile phones and automobiles Accelerate evolution and seize the technological dividends of large models.

To summarize, the large model is the end of this wheel The starting point of the AI ​​wave, computing hardware is the source of performance for the new concept of AI terminals to be implemented. Supporting the continued development of large models and AI terminals, Qualcomm has also contributed to the prosperity of terminal-side AI at the source.

The reason why we need to look for the source of end-side AI development today is because it is consistent with the needs of domestic consumer electronics manufacturers to seize the opportunities and technological trends of the times, work closely with the source, and advance hand in hand, which is conducive to obtaining the direction. Momentum for future growth.

At present, Qualcomm is one of the important AI forces that cooperates closely with domestic manufacturers.

On the one hand, Qualcomm and Snapdragon platforms are domestic Android Assisted by the globalization and high-endization of mobile phone manufacturers. Domestic leading brands are competing head-on with Apple in the high-end market. The close cooperation between Qualcomm and Snapdragon with domestic manufacturers has become a guarantee for the global competitiveness of domestic smartphone brands, allowing leading AI applications, AI mobile phones, and AI PC hardware to set benchmarks for the world. Currently, the first batch of flagship phones equipped with Snapdragon 8 Extreme Edition areFor example, Xiaomi, Honor, iQOO, etc. have established distinct market recognition in terms of cost performance, imaging capabilities, and gaming performance.

In addition, the growth of new energy vehicles is also expected to further progress.

At present, domestic new energy vehicles have established strong competitiveness in the global market. Seizing the opportunities of intelligence and achieving greater development can further consolidate their advantageous position. The Snapdragon automotive platform is equivalent to the cornerstone, supporting large-model-based intelligent driving and intelligent cockpit onboard vehicles, allowing new energy vehicles to maintain their competitive advantage and move more steadily and further in the global market.

For on-device AI, large models are the new A new generation of operating systems, and making good use of this new operating system requires full support from the ecosystem, tools, and resources. Through Snapdragon's software and hardware ecosystem, Qualcomm can provide ecological assistance for the development of on-device AI, help more developers make good use of large model technology, and promote the overall prosperity of AI products and applications.

To summarize, the market for terminal consumer electronics is vast, and intelligent upgrades have just begun. AI will sweep across all terminal categories we are familiar with. Standing at the forefront of the times, it is gradually becoming clear that Qualcomm and Snapdragon are the source of power.

Qualcomm and Snapdragon’s close embrace of AI has brought about the long-term coexistence of mobile computing and device-side AI. The story of device-side AI that has been running wildly for a year will continue to be told.