2023 is coming to an end, looking back on the technology development trend of this year, if you want to ask which technologies are the most out of the circle this year?The AI model must be the best.
From language understanding to image recognition, from intelligent recommendation to assisted driving, AI models are showing a strong influence in various fields. Especially for more consumers, generative AI applications are extending to consumer electronics such as mobile phones and PCs, and with its blessing, traditional hardware is undergoing a major upgrade of intelligence, which in turn has brought great changes to our lives and work.
1 Bringing generative AI to mobile phones and computers, where is the change?Such a statement is not an exaggeration. As AI becomes more and more creative, the way humans interact with each other is undergoing profound changes. A notable example is that when intelligent assistants integrate large, pre-trained models, smartphones are better able to understand and process human language, enabling more intelligent interactions. This means that the conversation between the user and the intelligent assistant will no longer be limited by the traditional command-based interaction, and will be more fluid and human, and the content of the conversation will be closer to the human.
In addition to enabling intelligent assistants to better understand and process human language, generative AI can also recommend corresponding applications, **, etc., according to the user's usage habits and preferences, so as to improve the user's user experienceand assist users to carry out more efficient task processing and improve user work efficiency.
While "moving" generative AI into mobile phones, PCs are not idle.
A number of computer manufacturers are actively embracing AI and jointly embracing the era of AI+PC.
With the blessing of generative AI, computers can better understand our creative needs, whether it is writing articles or doing **, and can provide us with personalized creative support. Even like smartphones, it can analyze the user's behavior and preferences to recommend suitable services and products for us, such as recommending a good movie, a good book, or the apps and services we need.
Whether it is to meet the needs of productivity or serve other aspects of life, the addition of AI will completely change the way and function of traditional mobile phones and computers, not only improve our work and life efficiency, but also open up new application scenarios and business models, and promote the development and innovation of the entire industry.
2. Go beyond the battle between software and hardware, and focus on the duel of AI capabilities.Just as whether a large meal is good or not depends on whether or not condiments are added. As a basic and core technology, AI has become indispensable in the process of improving the product experience, just like the condiments in delicious meals. This means that the competition between next-generation smartphones and PCs will not only be hardware and software, but also AI capabilities will become a decisive competitive factor, especially the capabilities of AI on the device side.
Compared with cloud AI, device-side AI has faster response time and lower latency, which is more suitable for those application scenarios that require real-time interaction and rapid response, such as smartphones, PCs, XR and other hardware platforms have become very suitable carriers.
In addition, device-side AI helps protect user privacy. In many cases, data can be processed and analyzed without leaving the user's device, making it more secure and reliable. Even in the absence of a stable network connection or poor network conditions, the AI on the device side can still run independently and provide users with basic services and functions.
Of course, device-side AI and cloud-based AI are not mutually replaceable, but complement and work together, and the two can work together to bring a more efficient, intelligent, and personalized user experience to smartphones, computers, and other devices. However, as one of the parts, the device-side AI will determine the final product experience, which needs to be driven by a powerful enough "power source".
There is no doubt that whether it is a smartphone or a PC device, the chip is the core and most important part. Similarly, chips play a decisive role in what kind of AI capabilities a device has.
As a global chip industry leader and an active participant in the AI field, Qualcomm plays a vital role in promoting the development and application of generative AI technologies. Taking the third-generation Snapdragon 8 mobile platform as an example, in order to improve its AI performance, Qualcomm not only upgraded the Hexagon NPU unit, but also equipped it with an independent power supply circuit, which not only brought a 98% performance improvement, but also reduced unnecessary wake-up and energy waste, resulting in a 40% increase in energy efficiency.
Based on the skyrocketing local AI computing power of the third-generation Snapdragon 8, it is possible for generative AI to run smoothly on mobile devices, and the third-generation Snapdragon 8 is the industry's first mobile platform that can run models with more than 10 billion parameters. And it is very fast, it only takes less than 1 second to generate an image through the AI Wensheng Graph model Stable Diffusion, compared to about 15 seconds for the previous generation.
On the PC side, Snapdragon also brought the Snapdragon X Elite, which is also equipped with an independent NPU with a computing power of 45 TOPS, which is enough to support generative AI models running with more than 13 billion parameters on the terminal side. In the StableDiffusion model test, AI painting was performed using a PC equipped with Snapdragon X Elite, and the image was produced in less than 1 second.
3 Qualcomm and terminal manufacturers, what experience can end-to-end generative AI bring?Having said all this, it will eventually return to the user itself, which is still the original question: with the support provided by Qualcomm on the device side AI, what practical application experience will generative AI bring?
Earlier, we mentioned the mobile assistant, which provides users with a convenient and fast way to interact, and with the improvement of AI computing power, the mobile phone intelligent assistant can not only achieve more accurate, smooth, and efficient conversations, but also have the ability to handle more complex tasks. For example, on the Xiaomi 14 series, based on the latest third-generation Snapdragon 8, Xiao Ai can understand and respond to the user's complex instructions, providing weather inquiry, news broadcasting, ** schedule management and other services.
Users can also ask Xiao Ai to help you create copywriting, write various plans and strategies, provide advice on clothing, food, housing and transportation, analyze hot topics and popularize various knowledge, and even let the voice assistant help you automatically convert a piece of text into **, the generation speed is faster, and the operation is not complicated and obvious, which is something that Xiao Ai could not do before, and the experience is not the same.
With the official release of OriginOS 4, the iQOO 12 series also has a new AI gameplay. I believe that many friends have a lot of ** stored in their mobile phones, and when they suddenly want to find a specific one**, they may look for it for a long time. If you are a user of iQOO 12, you can call up the blue heart small V, and say to it, "Little V help me find a portrait**" and the small V can quickly find what you want**.
The same method can also be used for files, especially those files that the user himself does not know where to store them, and Xiao V can also help you find them quickly.
Based on the powerful AI capabilities of the third-generation Snapdragon 8 mobile platform, even gaming phones have applied AI large model technology for the first time. The RedMagic 9 Pro is built into RedMagicOS9In 0, the game AI intelligent assistant can provide players with various help and support during the game, such as gameplay and game strategies, and can also automatically identify the content of the screen and make corresponding feedback, such as residual blood spraying, reloading, etc. Even when playing some FPS games, the virtual game assistant Mora will cheer the player on, as if there is a soft girl by the player's side.
Not long ago, Qualcomm also launched the most powerful PC platform in history, Snapdragon X Elite, whose powerful AI performance and excellent energy efficiency ratio also provide support for generative AI applications to land on the PC side, changing the way people interact with PCs. From AI expansion, intelligent creation to virtual character simulation, a series of new application scenarios are gradually emerging.
Taking AI expansion as an example, this technology AI can expand a picture according to equal proportions or free angles, and will expand other pictures according to the content of the image** and supplements. In fact, not only AI can not only intelligent** and supplement images, but can even generate ** based on text, **generation**, such as some AI intelligent production tools. Among them, Meta has launched Make-A-Video in 2022, which can use a given few words or a few lines of text to generate a short ** for a few seconds.
In the latest demo of the Snapdragon X Elite, laptops equipped with Snapdragon X Elite can even support entrepreneurs to complete more complex tasks such as creative design, budget analysis, business proposal production, and brand building, helping users achieve their business ideas and goals more efficiently.
And finally:
From natural language dialogue to copywriting** creation, from intelligent recommendation to personalized services, with the enrichment of application scenarios and the continuous development of AI technology, the application of generative AI in the next generation of smartphones and PCs will be more extensive and in-depth, and provide users with a more intelligent and convenient user experience. In general, the penetration of AI large models into various terminal devices has begun, and it is still in its infancy, and a bigger change is expected next year.
From another point of view, Qualcomm does play a key role in promoting "device-side AI" to the majority of users. At present, the shipment of products equipped with Qualcomm's AI engine has exceeded 2 billion, and its AI capabilities have widely penetrated into billions of edge terminal devices, which has not only successfully built a terminal AI ecosystem with "Snapdragon chips" as the core, but also spawned the implementation of a large number of generative AI application scenarios, and finally allowed more consumers to experience and participate in this AI-driven future transformation as soon as possible.
In addition to the CPU and GPU performance and energy efficiency improvements, the Snapdragon 8 Gen3's bigger upgrade is actually reflected in the support of AI models. In terms of AI, the Snapdragon 8 Gen3 processor adopts the latest Qualcomm AI engine, and introduces a new Hexagon NPU processor to replace the previous Hexagon DSP as the core of the AI engine, achieving up to 73Tops of total AI computing power, which is 98% higher than the previous generation and 40% more energy efficient. Snapdragon's skyrocketing local AI computing power in 8 Gen3 has made it possible for generative AI to run smoothly on mobile devices, opening a new era of generative AI for smartphones.
A prime example of this is the third-generation Snapdragon 8, which became the industry's first smartphone mobile platform that can run a 10 billion parameter model. This means that it can directly rely on local AI computing power to realize functions such as natural language dialogue and AI drawing without being connected to the Internet.
The Snapdragon X Elite is also equipped with a discrete NPU that supports INT4 and a computing power of 45 TOPS. If combined with CPU and GPU, the Snapdragon X Elite can achieve a total AI computing power of 75 tops, and can run generative AI models with more than 13 billion parameters on the device side.
In the Stable Diffusion model test, the Snapdragon X Elite is running Stable Diffusion 15 times only 75 seconds, compared to Intel's 207 seconds is greatly reduced. At the same time, the speed of stable diffusion optimized is increased to 06 seconds, a full 20 times improvement.