Image source: Generated by Unbounded AI

In the summer of 1985, 52-year-old Alvin Jacob and 6 friends crowded into a pizza shop In the offices upstairs, the mission of a newly formed company to provide "Quality Communications" is being discussed. At that time, what supported the operation of this new company was the founder's belief that "communication technology will definitely change human life."

After 39 years, Qualcomm has extended this belief to the field of terminal-side AI.

At the beginning of 2023, Qualcomm successfully ran the AI ​​Vincent graph large model - Stable Diffusion on the mobile phone side, verifying the possibility of the large model being implemented on mobile phones. Subsequently, mobile phone manufacturers released their own large models one after another, and shouted the slogan "The First Year of AI Mobile Phones" in early 2024.

Today, AI mobile phones have gone through three stages: calling APIs, accessing large models, and cooperating with chip manufacturers for implementation. Mobile phone manufacturers have already reached a consensus on AI "device-cloud collaboration". Under the idea of ​​using large models in the cloud to build capabilities and large models on the device to build intelligence, this fall’s mobile phone release season has become an AI creative competition.

The most impressive thing is Honor CEO Zhao Ming’s sentence of “order 2,000 cups of coffee.” The mobile phone smart phone that automatically places orders allows Luckin to sell orders in a radius of several kilometers. To improve user interaction experience, we have also seen different ideas provided by Xiaomi, OPPO, and vivo.

The power of on-device AI comes from the implementation of multi-modal AI, which can achieve mixed understanding of voice, text, and images. Only AI that can "read" the mobile phone screen and use the camera to "understand" the user's surrounding environment can achieve the ability to "operate for" the user.

On the other hand, it is the optimization and design of chip manufacturers from the SoC level.

The most thorough terminal-side mobile platform that embraces AI is Qualcomm’s Snapdragon 8 Extreme Edition this year. On this generation of chips, Qualcomm uses self-developed Oryon CPU based on its understanding of AI and terminal development needs. In response to the demand for AI data throughput on the terminal side, the Snapdragon 8 Extreme Edition provides quite "luxury" caches in the secondary CPU cluster, GPU, ISP, NPU and other units, which helps to improve the overall AI capabilities.

In fact, the changes in the overall architecture of this generation of mobile phone SoC are not the first time Qualcomm has used AI ideas to design chips. The previous generation’s third-generation Snapdragon 8 was Qualcomm’s “first mobile platform crafted specifically for generative AI.” This generation’s Snapdragon 8 Extreme Edition uses the second-generation customized Qualcomm Oryon CPU. Users It has the powerful performance required to handle complex AI tasks, such as dynamic multi-modal content generation.

On chips used in PCs (Snapdragon X Elite and X Plus), also relying on Oryon CPU and AI capabilities, Qualcomm is rapidly breaking through the X86 chip marketfield. In the automotive field, Qualcomm has also released cockpit and smart driving platforms with stronger AI capabilities, trying to combine the advantages of cockpit chips to break through the high-end smart driving chip market dominated by Nvidia.

“Qualcomm is developing from a company focusing on wireless connections to a company focusing on AI processing in the new era and making intelligent computing ubiquitous.”

As Qualcomm CEO Cristiano Amon) said at the 2024 Snapdragon Summit. Today's Qualcomm is no longer satisfied with its original communications dream.

Qualcomm is planning the next ten years of AI terminals by combining all types of terminal chips such as mobile phones, PCs, and automobiles.

How can a chip lead mobile phone AI innovation?

The competitive focus of mobile phone manufacturers has been entirely on AI capabilities. For flagship phones from the fall of 2024, almost every manufacturer emphasized their main AI functions at the beginning of the press conference.

An overall trend is that relying on the real-time interaction and multi-modal recognition capabilities of end-side AI, AI is reshaping the interaction logic of mobile phones, making the degree of "automation" higher and higher.

Mobile phone manufacturers that use the Snapdragon 8 Extreme Edition have more flexible AI interaction methods. For example, in Xiaomi's Super Xiaoai, users can ask Xiaoai to remember invoices and other information that appear on the screen at any time, and retrieve and automatically fill in the information when needed, instead of "search now". On Honor's side, the YOYO assistant has evolved into an intelligent agent, which can convert vague instructions into specific operations on the phone and "play" on the phone for the user.

The first batch of mobile phone products equipped with Snapdragon 8 Extreme Edition Source: Internet

The preference for different AI function designs is not only related to the differences in thinking of OEM manufacturers, but also to the chip manufacturers’ understanding of AI. The "mode and degree" of support also has considerable relevance.

For the Snapdragon 8 Extreme Edition, Qualcomm simply reconstructed the entire SoC using AI ideas.

The biggest change is the CPU design idea of ​​the Snapdragon 8 Extreme Edition, which has shifted from the three-cluster design of the previous generation to an architecture of two "ultra-large cores" plus six performance cores. This means that the Snapdragon 8 Extreme Edition, which removes the energy efficiency core, was designed to "maximize" the computing power from the beginning. On this basis, in order to support intensive computing, the Snapdragon 8 Extreme Edition is directly equipped with up to 12M shared cache for the two core clusters. You know, the total cache size of the performance cores of Apple A18pro and MediaTek Dimensity 9400 is only 2M and 6M. Based on this design, the Snapdragon 8 Extreme Edition has also achieved up to 45% performance improvement (single and multi-core) and 44% reduction in power consumption.

With such a magnitude of upgrade, Xiaomi Group partner, group president, and mobile phone department president Lu Weibing predicted in September that “2024 will be an inflection point for the chip industry. In the next month or so, bigEveryone will see the turning point appear! ”

Similar designs to the CPU are also present in other modules of the Snapdragon 8 Extreme Edition. For example, at the AI ​​acceleration level, Qualcomm has greatly improved the throughput of the NPU accelerator core, adding scalar and vector The number of accelerator cores is 70+ per second under common large models. tokens. On the other hand, in terms of mobile phone images, the ISP (Image Signal Processor) is also directly connected to the NPU, and the captured images are automatically optimized by the algorithm.

With the help of the AI ​​capabilities of the Snapdragon 8 Extreme Edition, the mobile phone Manufacturers have also achieved the effect of "AI everywhere" in the released products, Xiaomi 15 and iQOO. 13. Honor Magic7 and OnePlus 13 have implemented AI computational photography, AI optimized screen display and other functions.

However, simply making AI chips is not the end point of Qualcomm’s advancement of end-side AI. Siddhika Nevrekar, the company’s senior director of product management, said at the Snapdragon Summit, “Hardware alone cannot create powerful AI. This is like a racing car, it doesn’t help if there is no fuel in the tank.”

In order to make more consumers willing to experience device-side AI, Qualcomm has also started cooperation with software developers to jointly accelerate the release of Snapdragon 8 Extreme Edition’s AI capabilities.

Among them, Qualcomm is cooperating with large model manufacturers, such as Tencent Hunyuan. In the cooperation, Tencent Mobile Manager can identify fraud information with the help of the client-side model. In the cooperation with Zhipu, the client-side large model can use a variety of interactive methods, including using the camera for real-time voice conversations, uploading photos for conversations, and uploading videos. Have a conversation. On the other hand, at a more refined functional design level, Qualcomm has opened up Qualcomm AI to developers for free testing and deployment models. Hub accelerates the implementation of developers’ AI ideas.

Since Qualcomm verified the feasibility of large-scale AI models on the device, the trend of AI in mobile phones has become unstoppable, driven by Qualcomm and mobile phone manufacturers. , which has contributed to the current situation that the Android camp's AI is ahead of Apple's.

After decades of innovation in the field of high-performance and low-power computing, Qualcomm is extending such terminal-side AI capabilities to various terminal chips. .

From mobile phones, PCs, XR... to cars, Qualcomm all needs to be in the AI ​​terminal In this era, Qualcomm can be called the most active player.

On the one hand, Qualcomm has basically launched all kinds of smart hardware (such as VR, AR, wearable devices, robots, etc.) that have emerged in recent years. Corresponding chip platform. Many iconic smart terminal products that have caused a sensation in the market, such as Meta Quest. 2. Humane AI Pin and Ray-Ban Meta AI glasses are both based on the capabilities of Qualcomm chip platform

Another.On the other hand, Qualcomm is working hard to achieve breakthroughs in the process of AI transforming traditional terminals.

In the automotive field, Qualcomm has become a highly competitive smart cockpit supplier. According to data from Gasgoo, in the ranking of cockpit domain control chip brands in the Chinese market in the first three quarters of 2024, Qualcomm has a market share of 67.2%, far ahead of other traditional automotive chips and chip suppliers that also focus on high performance.

Cockpit domain control chip brand market share ranking in the Chinese market in the first three quarters of 2024 Source: Gasgoo

With its market share advantage, Qualcomm has also successfully occupied the market share of OEMs and consumers person’s cognition. Among them, after Jikrypton announced in July 2022 that it would upgrade the cockpit chip of Jikrypton 001 to Qualcomm Snapdragon 8155 for free, car sales began to surge. In just two months, Jikrypton 001’s monthly sales have exceeded the 10,000-unit mark from 5,000 units in July. Similarly, in the recent car cockpit chip upgrade plan promoted by Xpeng, car owners faced the "temptation" of Qualcomm Snapdragon 8295 chip and collected 4,000 crowdfunding quotas in just three days.

In addition to cars, in the process of PC AI upgrades, more and more PC manufacturers (such as Honor, Dell, HP, Lenovo, Samsung, Acer, Asus, etc.) have also simultaneously released mobile phones using Snapdragon Devices on the Dragon X Elite platform. Regarding the specific consumer experience, a Lenovo product manager in charge of the Yoga series told Guangcone Intelligence, “My own office notebook uses Qualcomm’s Snapdragon platform, and I have no problems running office software and watching videos. Moreover, the power consumption of the Snapdragon platform The control is very good, and the office life can easily last 8 hours.”

If we say that the success of AI chips in mobile phones comes from Qualcomm’s solid foundation in the mobile phone field. So in the fields of automobiles and PCs, what is Qualcomm relying on to make such rapid progress?

It is the application of AI, which fundamentally changes the interaction logic of smart terminals. Perhaps the large model in the cloud has greater capabilities, but it cannot achieve real-time interaction and privacy protection. Regarding the advantages of implementing end-side AI, Anmeng said, for example, “If generative AI can be implemented on mobile phones and PCs that understand the user better, then AI can rely on this personalized information to provide a faster response and more personality than the cloud. ”

It is precisely based on the idea of ​​AI terminal implementation, in order to make AI do more things, Qualcomm, which constantly iterates SoC AI performance, is favored by more and more terminal manufacturers.

On the PC side, on-device AI can greatly improve productivity. In the cooperation between Qualcomm and Microsoft, users only need to tell Copilot to refer to a specific document to complete PPT creation. Following this type of AI application, players in the PC industry have designed similar functions (such as Lenovo's vectorized storage of data, Honor's one-click search for local pictures and videos). While AI can understand pictures and documents, it can also rely on generative AI to perform secondary processing on the content locally.

In the automotive smart cockpit, it is reflected in the improvement of service capabilities by AI applications.

In the CarPlay era, the cockpit system can basically only control audio and video entertainment. Nowadays, in the "centralized" automotive EE architecture of various manufacturers, with the blessing of large models, smart cabins can not only respond to users' vague needs, but also automatically control windows, chassis, kinetic energy recovery and other electronics that affect car driving. Accessories provide a safer and more enjoyable driving experience.

In other words, the cockpit SoC equipped with end-side AI will have the opportunity to directly define the car driving experience. Furthermore, with the multi-modal understanding ability, it is possible to understand the AI ​​inside the car and also have the possibility to identify the situation outside the car.

“Vehicles capture layers of detail of their surroundings through sensors, transforming the real world into situations that humans and machines can digest and understand simultaneously. While driver assistance functions focus on identifying safety risks and driving functions, The same sensor can also identify situations in the user's journey. "

As Nakul Duggal, general manager of Qualcomm Technologies' Automotive, Industry Solutions and Cloud Group, said, with the self-developed Oryon. For CPU architecture, Qualcomm has also launched a new smart cabin chip and smart driving chip (Snapdragon Cockpit Extreme Edition platform and Snapdragon Ride Extreme Edition platform).

From Qualcomm’s actions in the fields of mobile phones, automobiles and PCs, we can feel that based on its experience in implementing terminal-side AI, Qualcomm is trying to equip all smart terminals with SoCs that emphasize AI capabilities.

As Anmeng concluded, "Generative AI will become the interface between people and applications. When end-side AI fully integrates the cloud and the terminal, we will see a new round of industry changes."

Qualcomm is fully embracing the AI ​​era.

Embrace the AI ​​era at the best time

"Qualcomm is In the process of transformation, everyone here is a witness.”

Anmeng believes that in an era when AI is subverting all previous smart terminals and communication technologies, the connections and interactions between various devices have become more complex. complex.

From wireless technology to fully embracing AI, it is a courageous decision for Qualcomm, which is about to turn 40.

Jack Ma once said in a speech that the best time for a company to change should be when the situation is at its best. “You must climb to the roof to repair the roof when the sun is shining, and never get down. If you go to the roof to repair it when it rains or snows, you may fall to your death on the roof." However, in reality, many large companies have not paid attention to such a simple sense of worry. Many former giants rose and fell with the times.

Qualcomm in the AI ​​era has successfully occupied the latest hot spot in AI terminals. Reflected in financial data, it is actually visiblerevenue growth.

According to Qualcomm’s fourth quarter report for fiscal year 2024 (as of September 2024), Qualcomm’s automotive business achieved revenue of US$899 million this quarter, a year-on-year increase of 68%. Although the current automotive business revenue still accounts for less than 10% of the total revenue, the expected continued growth rate (the company expects the automotive business to still grow by 50% in the next quarter) still implies that the automotive industry is the company’s most promising second sector. growth curve.

On the other side, the IoT business is mainly consumer electronics, edge network and industrial products. This quarter, the company benefited from the launch of new XR and AI PC products and the replenishment of inventory in the industry chain. This quarter, the company achieved a year-on-year growth of 21.7%.

Although, this performance is the first growth after six quarters. But combined with the fact that the just-released Rokid Glass adopts Qualcomm’s AR computing platform, Qualcomm has once again become the “only” reliable partner for manufacturers when AI terminals are about to explode.

At the just-concluded Qualcomm Investor Conference, Qualcomm set new financial goals for its QCT semiconductor chip business in the next five years, including that by fiscal year 2029, the fiscal year revenue of its automotive and Internet of Things businesses will reach 22 billion U.S. dollars; the fiscal year revenue of the automotive business increased to 8 billion U.S. dollars, and the fiscal year revenue of the Internet of Things business, including PC, XR, industrial related, etc., increased to 14 billion U.S. dollars.

When AI has become the consensus of the smart terminal industry, Qualcomm, as the leader of terminal-side AI innovation, has already planned its next decade in the AI ​​terminal era.