Shulou(Shulou.com)11/24 Report--

Some time ago, Qualcomm held the Snapdragon Summit in 2023. The most interesting features of this summit are the Snapdragon X Elite platform, the most powerful computing processor built for PC, and the release of the third generation Snapdragon 8 mobile platform. They are powerful products that combine terminal-side intelligence, top performance and energy efficiency.

Unlike in the past, after the release of Snapdragon's flagship platform, people focused on performance and power consumption. After the release of this year's Snapdragon X Elite platform and the third generation Snapdragon 8 mobile platform, people focused more on AI. In fact, these two products are both powerful platforms built for AI and enabled by AI. The third generation Snapdragon 8 mobile platform injects AI into the entire platform system and opens a new era of generative AI.

As we all know, the influence of AI on people's production and life is continuously expanding and deepening with the naked eye, especially with the hot and rapid application of generative AI as the key node. Its development has further improved the way of human-computer interaction, especially terminal-side generated AI is essential for creating powerful, fast, personalized, efficient, secure and highly optimized experiences, and these transformative experiences will be driven by AI-enabled chips at the bottom.

For Qualcomm, all this is not achieved overnight, but a natural result after years of deep ploughing.

After years of deep ploughing and evolution, AI has become the key basic capability of Snapdragon platform. Qualcomm's friends who follow Qualcomm believe that in the past years, Qualcomm has been emphasizing the AI capability of Snapdragon Mobile platform, while AI is playing a more and more important role in Snapdragon Mobile platform.

But no matter how the technology evolves and the application is rich, generally speaking, the direction they always adhere to is to apply the artificial intelligence technology with powerful computing power to the terminal to promote the development of terminal-side AI.

For example, as early as 2007, Qualcomm launched its first artificial intelligence project. At that time, it was still the world of functional phones. After nearly 10 years of accumulation and achievements, Qualcomm formally integrated the first generation artificial intelligence engine Zeroth on the Snapdragon 820 processor in 2015, and realized the running neural network on CPU.

At that time, the AI application mainly realized some basic functions, such as intelligently classifying according to the photos taken by users, identifying the images of objects and cars, and storing them in two different folders to facilitate users to better manage the photos.

In 2016, Qualcomm released the Snapdragon nerve processing engine SDK, and in early 2017, the second generation artificial intelligence engine was launched with the release of Snapdragon 835. After the introduction of this artificial intelligence chip, smart phones began to enter the era of intelligent photo album management and face recognition, especially the application of face recognition, which makes the existence of AI in intelligent terminals to a new level.

In 2018, with the release of the Snapdragon 845 mobile platform, Qualcomm's third generation of artificial intelligence engine AI Engine came out. Qualcomm began to introduce heterogeneous parallel computing of CPU, GPU and DSP in this generation of artificial intelligence engines. The performance of AI computing is three times higher than that of the previous generation, which enables smartphones to have powerful top-level AI computing power support.

In 2019, AI Engine came to the fourth generation with the Snapdragon 855 mobile platform, which was able to achieve more than 7 trillion operations per second, making AI have more diverse applications on smartphones. For example, Qualcomm and NetEase worked together to use AI Engine to accelerate the real scene AR translation function on some Snapdragon mobile platforms, and they cooperated with Tencent to carry out terminal-side practical applications with the help of Snapdragon nerve processing SDK. For example, Mobile QQ's "high-energy dance studio" function and so on.

The performance of the fifth generation AI Engine,AI on Snapdragon 865 mobile platform in 2020 has more than tripled compared with the previous generation, reaching 15TOPS. With the help of powerful computing power, the application of AI in intelligent terminals has also been extended to games, photos, interaction and other fields. For example, in the image field, the fifth generation AI Engine acts on Spectra 480 ISP through its powerful AI ability to optimize the image digital information, such as noise removal, highlight suppression, dim light compensation, and so on, so that the photo performance of many flagship mobile phones at that time, especially the dynamic range performance has been qualitatively improved.

Next, in the Snapdragon 888 mobile platform, the sixth generation AI Engine continues to evolve in performance, which also promotes the smart terminal to experience more and more excellent in images, games and so on. For example, Qualcomm has worked with Snapchat to use the AI Model Enhancement Kit to quantify a series of AI lens models to improve the accuracy and performance of face recognition, and to achieve portrait matting in video, and so on.

The seventh generation AI Engine of Snapdragon 8 Gen 1 mobile platform in 2021 doubled its performance and energy efficiency, while focusing on bringing smarter sensing capabilities to terminal cameras, such as better autofocus, auto exposure and automatic white balance (3A) and details for face detection, and when the terminal senses that someone is peeking over the user's shoulder, the terminal will automatically lock the screen. At the same time, in Snapdragon 8 Gen 1, Qualcomm also integrates AI technology into Spectra ISP, modem and RF system to achieve real-time face recognition, AI antenna tuning technology and other capabilities.

And the previous generation of Snapdragon 8 Gen 2 platform is AI everywhere, no matter the experience of image, audio, connection or game, there is the deep participation of AI. At the same time, "Cognitive AI" has become a new direction and stage that Qualcomm is committed to promoting, so that AI can understand human cognitive emotion from the dimension of context and timeline, which is undoubtedly a milestone.

Through the review of the evolution of AI technology and applications on Snapdragon platform, we can find that AI does play a more and more critical and diverse role in the terminal experience, from the role that only occasionally appears in the image, context awareness and other corners in the past, to gradually develop through all the key areas of the entire chip, and now it has become the most critical basic capability.

The third generation Snapdragon 8 and Snapdragon X Elite announced that the AI era of mobile chips has come. In the past, we talked about "terminal experience". Most people still focus on CPU performance and have a broader field of vision. Many people will also focus on GPU, ISP, audio, 5G, Wi-Fi and other capabilities, which indeed constitute a more comprehensive terminal experience. The concept of integrating the six pillars of connection, AI, video, audio, games and security put forward by Qualcomm before has brought the Snapdragon mobile platform an omni-directional mobile experience with few flaws and shortcomings. what can not be ignored is that behind the six pillar experiences, there is a key underlying driving force, that is, AI.

This 2023 Snapdragon Summit shows us that AI is changing from the driving force at the bottom to the focus of attention. On the Snapdragon 8 Gen 3 (the third generation Snapdragon 8) and Snapdragon X Elite platforms, Qualcomm thoroughly focused their technical leadership and leadership in terminal-side AI on the two chip platforms, making AI truly the "star" of a comprehensive terminal experience change.

For example, on the third generation Snapdragon 8 mobile platform, AI Engine (Qualcomm AI engine) has developed to the ninth generation, and has made positive innovations. Qualcomm upgraded the past Hexagon tensor accelerator to Hexagon NPU, so that it can coordinate the entire platform. At the same time, Hexagon NPU itself has upgraded its new micro-architecture, with a 98% improvement in overall performance and a 40% improvement in energy efficiency.

At the same time, Qualcomm in Snapdragon 8 Gen 3 adds a directly connected channel between Hexagon NPU vector unit and memory, which greatly enhances the processing efficiency of AI operation. At the same time, the sensor hub has also been upgraded, with memory, two miniature NPU, Spectra DSP, double ISP and other different modules, which can provide efficient AI operation and processing anytime, anywhere.

With these powerful upgrades, Snapdragon 8 Gen 3 can support a variety of generative AI models, including Meta Llama 2, and has been able to implement large models with running parameters of more than 10 billion and can execute up to 20 Token per second. By running these powerful large models on the terminal side, users no longer have to rely entirely on the cloud.

Earlier this year, Qualcomm used Stable Diffusion for the first time in the world to demonstrate the terminal-side deployment technology of image generation, which runs entirely on the terminal side and takes no more than 15 seconds. The powerful AI boost can be seen by running this model in only about half a second through the third generation Snapdragon 8.

In terms of image, Snapdragon 8 Gen 3 integrates three perceptual ISP. With the powerful AI capability, Snapdragon 8 can now support up to 12 layers of semantic segmentation, which can identify objects and scenes more accurately.

At the same time, it also supports intelligent elimination of objects in video scenes, and the super night view function is no longer limited to photography, but also can be used for video recording, both of which are difficult to achieve without strong AI support.

Even 5G incorporates the power of AI. Snapdragon 8 Gen 3 integrates the world's first modem Snapdragon X75, which combines millimeter wave hardware (QTM565) and Sub-6 hardware, and also has the first dedicated hardware tensor accelerator. Qualcomm 5G AI processor II increases AI performance by 2.5 times, so that you can choose the best frequency more intelligently to achieve the best connection.

The new Snapdragon X Elite computing platform is equipped with heterogeneous Qualcomm AI engine, which can realize the computing power of 75TOPS through NPU, CPU and GPU, and its AI performance is 100 times higher than that of six years ago.

Among them, the Hexagon NPU computing power of the new Snapdragon X Elite computing platform can reach 45 TOPS. When running the 7 billion-parameter Llama 2 model on the notebook, it can provide fast and accurate response, and the processing speed can reach 30 Tokens / s.

Not only that, it can also completely run a generative model with more than 13 billion parameters on the terminal without the need for cloud resources. In this way, generative AI can help you draft complete emails, transcribe meeting minutes, or quickly discuss them with the help of text generation and image generation tools. Based on the operation and development of the parameters this year, it is believed that the capability may be further updated when the commercial machine is listed in the future.

These things used to be intelligently implemented in data centers, but now with Snapdragon X Elite computing platform, the same thing can be done locally in PC. This is based on Qualcomm's years of experience in heterogeneous architecture design, which makes it the perfect solution for modern AI PC.

It can be seen that whether it is the third generation Snapdragon 8 or the new Snapdragon X Elite computing platform, they are all chip platforms designed for AI or even defined by AI, and under the definition of AI, their connectivity, video, audio, games, security and other pillar experiences have been gratifyingly enhanced.

The "AI era" of mobile chip platforms has come.

The last kilometer of the landing of the AI era, of course, it is not enough for the chip platform itself to enter the "AI era". For the majority of users, these AI capabilities must be transformed into actual products and applications before we can really enter the AI era. These are inseparable from the close cooperation between Qualcomm and developers at the software and tool level, and even with ecological partners.

For example, for the landing of the mobile generation AI capability of the third generation Snapdragon 8 platform, Qualcomm has a leading software stack product specifically for edge-side AI, Qualcomm AI software stack, which can support model optimization from the software level. Through a series of tools such as algorithm, model development, AI model plug-in kit, compiler and AI engine Direct in Qualcomm AI software stack, developers can easily introduce the AI functions and experiences they want to achieve into the terminal.

At the terminal and application level, Qualcomm also has in-depth cooperation with partners. For example, Xiaomi, as a terminal manufacturer, has long defined the strategy of "AI full enablement". On the AI model, Xiaomi chose to focus their large model on end-to-side computing to optimize it. At this Snapdragon Summit, they demonstrated that the self-developed 6 billion-parameter model (6B model) runs smoothly on the Snapdragon 8 Gen 3 platform. According to Xiaomi, they run on a special NPU, and the instruction of 1000 tokens only takes 2.2 seconds to generate the first word, which is the result of the deep cooperation between Xiaomi and Qualcomm.

In addition to terminal hardware vendors, Qualcomm is also working with software manufacturers to promote the landing of generative AI applications. For example, they and domestic HuiCarp Technology have developed a neural network for the third generation Snapdragon 8, which can reconstruct the missing parts of photos, namely "photo expansion". It allows users to zoom photos to make them look wide-angle, even if they are not shot with a wide-angle lens.

As for the Snapdragon X Elite computing platform, the most important partner of Qualcomm in the PC field is Microsoft, which has been working with the Microsoft engineering team to optimize the features of the new platform. Microsoft said that the vast majority of top applications will run natively or through seamless simulation on Windows PC with Snapdragon X Elite computing platform at ultra-fast speed and energy efficiency, especially Microsoft's own productivity applications. Including Word, Excel, Powerpoint, Edge, Teams, OneDrive, OneNote and Outlook are native.

For Windows 11, Microsoft improved Windows Studio Effects, announced the launch of AI Library, introduced generative AI in updates for inbox applications such as Copilot and Paint, and partnered with Camo, Luminar Neo, WhatsApp, etc., to optimize the generative AI application experience on the Windows platform, all of which will be experienced on Windows PC driven by Snapdragon X Elite computing platform.

In addition, Qualcomm has partnered with Microsoft to launch Windows Dev Kit 2023 for developers. This toolkit contains everything you need to develop Windows applications for Snapdragon platform, all on one device, including some excellent developer tools, such as Visual Studio and Unity runtime engine, which are natively ported to the Arm platform. There are also ONNX Runtime and related execution providers. All these will help developers to create native applications with powerful AI capabilities.

In conclusion, I believe that with the efforts of Qualcomm and its partners, not only the mobile chip platform will move towards the AI era, but more importantly, all aspects of our mobile life will become better because of our integration with AI.

Especially at present, the process of digital intelligence transformation in various industries is advancing at full speed, and Qualcomm, which stands at the intersection of a number of key technologies and industry trends, will surely rise abruptly based on accumulated strength, promote the integrated development and innovation of AI together with the whole industry, change the way we experience the world, and make the future of digital intelligence a reality.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.