Written by | Neocortex Group
Edit | Jeff Wang
This week is earnings week for major U.S. tech companies. The top three companies in the field of generative AI, Microsoft, Alphabet (Google's parent company), and Meta (Facebook's parent company) all delivered earnings reports that exceeded Wall Street's expectations.
At the end of July, when we analyzed the earnings for that quarter, the business logic of the old world was still solid, and both Alphabet and Meta had won the approval of the capital markets with their excellent advertising growth. However, at the end of October, generative AI gradually began to change that, at least in the eyes of Wall Street, where the weight of advertising has been declining and it has gradually become a business in the past tense, and the business in the ongoing and future tense is AI.
This change is directly reflected in the fact that after the earnings report, Alphabet's stock price plunged 9.5%, and its market value evaporated by more than $166 billion in a single day, the largest decline since March 2020. Considering that the new crown epidemic broke out globally that month, and there were 4 circuit breakers in the US stock market in 10 days, Alphabet was just a fish in the pond that was affected, and it was not representative. Therefore, this plunge is the worst defeat in Alphabet's history. However, looking back at this earnings report, at least in terms of revenue, Alphabet did not live up to Wall Street's expectations, with the company's revenue of $76.7 billion in the quarter, higher than analysts' expectations of $76 billion.
The key issue is still in the cloud computing business, where Google Cloud's revenue grew 22% in the third quarter, the lowest growth rate in at least 11 quarters. As a third-place catch-up in the cloud computing market (behind Amazon and Microsoft), the slowdown is obviously bad news. Conversely, rival Microsoft's cloud computing product, Azure, is accelerating its revenue, rising 29% in the quarter, 3 percentage points higher than the previous quarter. Microsoft said that the three-point increase came from the hot sales of generative AI products, with the number of customers of OpenAI's GPT-4 model increasing by 63% to 18,000 in the quarter.
If Microsoft's regain of the trust of the capital markets lies in the fact that it has found a business value for generative AI – cloud computing, and Alphabet's failure is that it has not yet seized the opportunity, then the short-term drop in Meta's stock price reflects the fact that the company has not been able to fully convince investors of how generative AI can realize business value on a wide range of social apps, including Facebook, Instagram, WhatsApp, etc. Zuckerberg's idea is that generative AI can bring more fun to users, and the other is that generative AI can be used as a business assistant, and both of these paths to commercialization seem to be particularly tortuous.
Generative AI is a good thing, but how to cover the high cost of generative AI is the next problem that every tech company needs to solve: either directly let generative AI services generate revenue, such as cloud computing, or let generative AI increase the stickiness of hardware or software to expand the user base and ultimately indirectly generate revenue. At the moment, it seems that Apple should choose the latter, and Mark Gurman mentioned in a leak that three Apple senior vice presidents are integrating generative AI into each of Apple's products, including smarter Siri assistants, better apps like Apple Music, Pages, and Keynote, and more convenient developer tools.
On top of that, the impact of generative AI has also spilled over into a field that has long been overlooked by the mainstream – the personal computer. This week, Qualcomm released the Snapdragon X Elite chip for Windows laptops, which can run generative AI models involving more than 13 billion parameters and has 4.5 times the AI processing power of other laptop chips. On the same day, it was revealed that AMD and Nvidia had also joined the race, and their PC chips would be launched in 2025.
According to IDC, since the M-series chips were added to the Mac lineup, Apple's share of the PC market has almost doubled in three years. What's more, more and more developers are finding that the high performance and large memory of M-series chips are particularly good at running generative AI locally. Whether it is Qualcomm, AMD, or NVIDIA, they have been pulled into the same front by Microsoft, and their goal is to make generative AI run better and faster on personal computers, and to a certain extent, break through the limitation of computing power cost bottlenecks on the popularization of generative AI.
You can pay attention to YiMagzine's latest IP "NewNewThing".
The following is the most noteworthy smart news of the past week, compiled by the new IP for everyone, please enjoy.
Key Points
Financial report
Microsoft's growth in the third quarter exceeded expectations, and AI-driven effects began to appear;
Alphabet's third-quarter ad sales were strong, but the cloud computing business disappointed;
Meta's third-quarter results beat expectations, but Quest lost more.
Models and applications
Apple is moving forward with a massive generative AI catch-up program;
AI makes Google Maps a search engine for local life;
Boston Dynamics has connected ChatGPT to the robot dog Spot.
Chips
Nvidia and AMD will produce computer chips based on the Arm architecture;
Qualcomm Releases Mobile Chip Supporting Generative AI.
U.S. AI chip export controls came into effect on October 23.
Financial report
Microsoft's growth in the third quarter exceeded expectations, and the AI-driven effect began to appear
On Tuesday, Microsoft announced its results for the third quarter ended September 30, 2023 ("Q3"), the first quarter of fiscal 2024, with all business results beating Wall Street expectations, mainly due to the strong performance of its cloud computing and generative AI. As a result, Microsoft's stock price jumped 6% in after-hours trading. In the past three quarters, Microsoft's stock price has fallen by 6% overall, and this good performance has allowed Microsoft to regain the trust of the capital market.
Microsoft's earnings beat expectations
In the third quarter, Microsoft's revenue was $56.5 billion, up 13% year-on-year, and net profit was $22.3 billion, up 27% year-over-year. Profit growth on revenue is back in the double digits and is accelerating from the previous quarter, beating expectations from Wall Street analysts and Microsoft itself.
In terms of business:
· Microsoft's intelligent cloud business contributed $24.26 billion in revenue, up 19% year-over-year. The segment includes Azure Public Cloud, SQL Server, Windows Server, Visual Studio, Nuance, GitHub, and Enterprise Services.
· Productivity and business process revenue was $18.59 billion, up 13% year-over-year. The segment includes Microsoft 365 productivity app subscriptions, LinkedIn and Dynamics enterprise software. In a conference call with analysts, Nadella said that Microsoft's corporate messaging tool Teams, similar to DingTalk or Feishu, now has more than 320 million monthly active users, up from 300 million six months ago.
· The personal computing business, which includes Windows, Xbox, Bing and Surface, posted revenue of $13.67 billion, reversing the decline in the previous quarter and increasing 3% year-over-year.
Among all businesses, the Intelligent Cloud division excelled. In particular, the division's flagship cloud computing product, Azure, saw revenue increase of 29%, up from 26% in the previous quarter. Microsoft said about 3 percent of the increase came from generative AI products, such as OpenAI's GPT-4 language model access, which now has 18,000 customers, up from 11,000 in the previous quarter.
Microsoft continues to invest in AI
Microsoft has become a frontrunner in the field of artificial intelligence and has pledged to actively invest in artificial intelligence to meet demand. In this earnings report, Microsoft's capital expenditure in the third quarter was $11.2 billion, up from $10.7 billion in the previous quarter and the largest expenditure since fiscal 2016. In response, Microsoft executives said that this figure is likely to grow in the rest of the fiscal year, and the company's capital expenditure will exceed $44 billion as a result.
Notable among them is Microsoft's $13 billion investment in startup OpenAI, the maker of the popular generative AI chatbot ChatGPT. Microsoft is currently improving its product suite, including Office, Windows, search engine, and security software, to add features based on OpenAI technology to attract more enterprise customers who want to use ChatGPT and other generative AI in their applications.
In an effort to boost ad sales and take market share from Google, Microsoft has added ChatGPT technology to the Bing search engine. In this earnings report, search and news ad sales increased by 10% in the quarter.
At the same time, sales of Microsoft's commercial cloud subscriptions, which include Excel, Word and Teams, accelerated, up 18% in the quarter. The integration of Microsoft's generative AI assistant Copilot with these products will be widely available to commercial customers starting next month for an additional $30 per month on top of their existing subscriptions.
In Monday's annual report, Microsoft CEO Nadella said in his annual letter that generative AI is the fifth major shift Microsoft faces after PC and Windows, web, mobile, and cloud computing. Microsoft has to keep up with this shift, all-in AI.
Reference Links
https://www.microsoft.com/en-us/Investor/earnings/FY-2024-Q1/press-release-webcast
Alphabet's ad sales were strong in the third quarter, but the cloud computing business disappointed
On October 24, Alphabet (Google's parent company) announced its financial results for the third quarter of fiscal year 2023 ended September 30, with third-quarter revenue of $76.7 billion, an increase of 11% year-on-year, slightly higher than analysts' expectations of $76 billion; Operating profit was US$21.34 billion, an increase of more than 24% year-on-year.
Alphabet's performance exceeded Wall Street's consensus expectations, but the cloud computing business performed poorly, with growth falling to its lowest level in at least 11 quarters, causing Alphabet's stock price to plunge 5.7% in over-the-counter trading after the U.S. stock market closed.
Performance specifics
Alphabet's revenue is mainly composed of three parts: Google services, Google cloud and other businesses.
· Google's services include advertising, paid memberships and hardware, with advertising accounting for about 90% of revenue. In the third quarter, Google's ad revenue increased 9.5% year-over-year to $59.65 billion, accelerating from the 3% increase in the previous quarter. At the same time, profits soared 42% to $19.7 billion. Among them, Google's search and other advertising revenue increased by more than 11% year-on-year to $44.03 billion. YouTube ad revenue was $7.95 billion, up 12.5%. The video streaming platform returned to growth in the last quarter after three consecutive quarters of decline.
· Google's cloud business revenue rose 22% to $8.4 billion in the third quarter, the slowest growth since at least the first quarter of 2021, after Wall Street expected Google Cloud to bring in $8.62 billion in revenue. As a third-place catch-up in the cloud computing market (behind Amazon and Microsoft), the slowdown is obviously bad news. What's more, the revenue growth of Microsoft's Azure cloud business is still accelerating, reaching 29% in the quarter, and the gap between the two sides is still widening.
· Alphabet's other businesses, including Waymo, a self-driving car project, and Verily, a life sciences unit, brought in $297 million in revenue but lost $1.2 billion, in line with analysts' forecasts.
What else did the earnings meeting convey?
· Significant layoffs and cost reductions: Alphabet had 182381 employees as of September 30, compared to about 119,000 at the end of 2019. In January, Alphabet laid off about 12,000 employees, or about 6% of its global workforce, to adapt to "different economic realities". Alphabet said severance and related expenses were $2.1 billion in the first nine months of the year.
· Increased investment in AI: Alphabet said it expects "higher levels of investment" this year and next, with the "vast majority" of capital expenditure in the third quarter coming from investments in AI-enabled infrastructure such as servers and data centers. From the beginning of the year to Sept. 30, Alphabet's R&D spending increased 14% to $33.3 billion. Ruth Porat, president of Alphabet, said, "We remain focused on a lasting redesign of our cost base to create investment capacity to support our growth priorities, the most important of which is artificial intelligence".
· Gemini is followed by a large model: there have been several media reports that Google's most advanced large model, Gemini, has been tested on a small scale and will be officially launched by the end of this year. And in the earnings call, Alphabet CEO Sundar Pichai said that in addition to Gemini, Google also plans to launch other large models in 2024.
· Inserting ads into Google's new searches: Advertising accounts for about 90% of revenue and continues to generate the majority of Google's revenue, so Google wants to increase revenue by adding ads to Google's new search (SEG) searches. Pichai said Google is planning to experiment with a native ad format for Google's new search, using generative AI to create relevant, high-quality ads and deliver tailored ads at every step of the search process.
Reference Links
https://abc.xyz/assets/4a/3e/3e08902c4a45b5cf530e267cf818/2023q3-alphabet-earnings-release.pdf
Meta's third-quarter results beat expectations, but Quest's losses were even worse
On Wednesday, Meta announced its results for the third quarter ended September 30, 2023 (hereinafter referred to as the "third quarter"), with revenue and profit growth exceeding expectations, with revenue achieving the fastest growth rate since September 2021.
Despite Meta's strong earnings growth this time, Meta CFO Susan Lee expressed concern about weak advertising demand so far in the fourth quarter during the earnings call, and Meta's shares initially rose about 4% in after-hours trading, before retreating and falling about 3%.
Specific performance
Meta's revenue in the third quarter was $34.2 billion, higher than Wall Street's consensus forecast of $33.5 billion, a year-on-year increase of 23%; Net profit reached $11.6 billion, up 164% year-on-year, well above the consensus estimate of $9.4 billion.
Meta's revenue is mainly composed of Family of Apps (its app family) and Reality Labs.
· The Family of Apps business, which includes products such as Facebook, Instagram, Messenger and WhatsApp, generated $33.6 billion in total advertising revenue in the third quarter, up 24% year-over-year. At the same time, the use of these apps grew strongly, with "daily active users" up 7% year-on-year to 3.14 billion.
· Reality Labs, which primarily includes virtual reality and augmented reality hardware, software, and content, saw its third-quarter revenue fall 26% to $210 million. The division lost $3.75 billion in the third quarter, compared with a loss of $3.67 billion in the same period last year. So far this year, Reality Labs has posted a loss of $11.47 billion.
What else did the earnings meeting convey?
· Prioritizing AI: Zuckerberg said AI will be the area where Meta invests the most in engineering and computing resources in 2024. Artificial intelligence has been, and will continue to be, an important part of Meta's efforts to improve efficiency and reduce costs.
· Stick to Reality Labs: Susan Lee said the results of Reality Labs will be applied to more other applications. For example, Meta launched Avatar, a virtual digital human service, and users have created more than 1 billion avatar images. Another example is the Ray-Ban Meta smart glasses, which can help users use and share Meta apps and even support live streaming functions, helping to create a more engaging content ecosystem.
· Increased hiring: Zuckerberg said Meta plans to hire more AI-focused technologists for this purpose, and Meta expects payroll to increase next year, with a particular focus on "higher-cost tech positions." Susan Lee said the number of employees at the end of 2024 should be "significantly higher" than current levels, after which growth will slow.
· Weak ad demand: Susan Lee also mentioned that Meta expects a good fourth-quarter result, but warned that final revenue could fluctuate due to the current situation in the Middle East.
Reference Links
https://investor.fb.com/investor-events/event-details/2023/Q3-2023-Earnings/default.aspx
Models and applications
Apple is "very anxious" about generative AI and is pushing ahead with a massive catch-up program
On October 22, Mark Gurman wrote that Apple is "very anxious" about OpenAI and other companies taking the lead in generative AI, and is currently launching a huge program to try to catch up, with three senior vice presidents involved.
What does Apple want to do, and what is the current progress?
What Apple wants to do is not only to develop individual AI products, but also to better improve the development and application ecology of Apple's apps with the help of AI. The overall plan is led by John Giannandrea and Craig Federighi, Apple's two senior vice presidents of AI and software engineering, with the participation of Eddy Cue, senior vice president of services, and has a planned budget of $1 billion per year.
· John Giannandrea is currently responsible for developing the underlying technology of the new AI system and improving Siri, which is expected to be loaded with generative AI and may be launched next year.
· Craig Federighi is working on adding AI to the next iOS release, and is also working on how to integrate generative AI into Xcode-like development tools to make app developers in the Apple ecosystem work more efficiently.
· Eddy Cue is trying to embed AI capabilities into as many apps as possible, including Apple's own Apple Music and productivity tools such as Pages and Keynote.
Apple has not yet finalized whether the next step will be to deploy generative AI entirely on-device or fully in the cloud, or possibly a combination of both.
A summary of Apple's AI development disclosures
In July this year, Bloomberg reported that Apple has built its own AI framework, Ajax, as well as its own chatbot, AppleGPT, internally. Both projects were founded in 2022, Ajax is built on Google Jax, the machine learning framework of Google Cloud, and AppleGPT is also similar to similar applications such as ChatGPT, but does not reflect more prominent features.
In September, more information about Apple's AI base model team was revealed. The team is led by Pang Ruoming, a Chinese scholar who previously worked at Google Brain, with a total of 16 people, almost all of the key characters from Google, and the team costs millions of dollars a day in research and development. In addition, Apple has a visual intelligence team that is focusing on AI that generates images, videos, and 3D scenes, similar to Midjourney or OpenAI's Dall· E 2。
Apple plans to hold a special AI product launch next year.
Reference Links
https://www.bloomberg.com/news/newsletters/2023-10-22/what-is-apple-doing-in-ai-revamping-siri-search-apple-music-and-other-apps-lo1ffr7p?srnd=undefined
AI makes Google Maps a search engine for local life
On October 26, Google Maps announced a series of new updates. In addition to making navigation easier, the app also offers a new way to search for more inspiration. After the update, Google Maps is more like a search engine for local life.
What are the updates?
· Immersive Route View: At this year's I/O Developer Conference, Google introduced the Immersive Route View feature, which allows users to preview a 3D panorama of their trip, including visual walking navigation, time sliders, simulated traffic and weather conditions, and more. With this update, Google has expanded the availability of the feature to 16 locations including Amsterdam and Barcelona.
· Lens function: This function uses AI and AR technology to help users better understand their surroundings and provide information such as nearby ATMs, transportation stops, restaurants, and coffee shops. The feature has also been expanded to more cities, including Austin, Las Vegas, Rome, São Paulo, and Chinese Taipei.
· More detailed navigation maps: Navigation maps will more accurately reflect the real world, including more realistic representations of buildings and improved lane details. For U.S. users, it will display High Rider (HOV) lane information.
· EV-related information: The app will provide EV drivers with more information about charging stations, including charger compatibility and charging speeds. Users can also check the last time of use of the charging station to avoid damaged charging stations.
New search methods: Users can enter keywords to get photo-based search results and location information.
Google Maps is more than just searching for maps
AI makeovers are making Google Maps more and more like a search engine. When you don't know exactly what you want to do or where to go, Google Maps can give you more suggestions.
If you enter keywords such as "animal latte latte" and "autumn maple leaf", the map will feedback images and location information related to the entered text; If you're in Tokyo, type in "things to do" and Google Maps will give you suggestions for "anime", "cherry blossoms" or "art exhibitions", as well as the location of the event.
Miriam Daniel, head of the Google Maps team, said that Google analyzes photos uploaded by billions of users to match the text they enter with relevant images, and provides image location information, "We want to provide more travel inspiration when users need it".
Reference Links
https://blog.google/products/maps/google-maps-october-2023-update/
https://www.theverge.com/2023/10/26/23932315/google-maps-ai-immersive-view-ev-charging-search
Boston Dynamics has connected ChatGPT to the robot dog Spot
On Oct. 26, Boston Dynamics uploaded a video on YouTube of its four-legged robot dog, Spot, wearing a top hat and a mustache, able to interact with employees in real time and lead them around the company's facilities as a robot tour guide.
What are the capabilities of Spot the robot dog?
Spot is a power-balanced quadruped robot designed by Boston Dynamics that uses four mechanical legs to move and is used in a wide variety of research applications and industries, including remote sensing and inspection, building monitoring, field documentation, radiation detection, environmental monitoring, and more. Currently, Spot has mapping, localization, and autonomous traversal systems, collectively known as GraphNav. With the GraphNav system, Spot is able to autonomously adapt to different types of terrain, and operators can access autonomous navigation using the auto-walk feature on the robot controller tablet.
How does this robot dog guide work?
To enable Spot to "speak," Boston Dynamics uses OpenAI's ChatGPT API as well as some open-source large language models (LLMs) to train its responses.
· Matt Klingensmith, chief software engineer at Boston Dynamics, said the team tagged the location of the building where Spot was located with a short description and fed the 3D building map data to the model, including ChatGPT.
· Subsequently, combined with a positioning system, the robot combines 3D building map data with images acquired through the gripper and cameras on the body, enabling it to "get more information about what is being seen before generating a response". Finally, the LLM is invoked to generate a content response.
· In addition, to allow Spot to interact with its audience and environment, Boston Dynamics integrated Visual Question Answering (VQA) and speech-to-text software that enables Spot to add captions to images and answer questions about images.
· In terms of hardware modification, Spot is equipped with a ring array microphone with LEDs, as well as some default body language and external decorations to enhance the sense of interaction with the audience.
Explore the intersection of AI and robotics
The company also noted that there are still some cases in R&D where models make up answers. Still, Matt Klingensmith says robots offer an excellent way to "land" large base models in the real world. Large language models such as ChatGPT can provide cultural context, common sense knowledge, and flexibility, which is useful for many bot tasks – for example, the ability to assign tasks to bots just by talking to them will help reduce the amount of learning using these systems.
Boston Dynamics says that by incorporating large language models, robots will perform better when working with or around humans — whether as tools, guides, companions, or entertainers.
Reference Links
https://bostondynamics.com/blog/robots-that-can-chat/
Chips
Nvidia and AMD will produce computer chips based on the Arm architecture
On October 23, it was reported that Nvidia and AMD are using the Arm architecture to develop CPUs, and the new chips are planned to be used to run Microsoft's Windows operating system. Previously, only Qualcomm was developing Arm-based PC chips for Microsoft, but now all chip giants are involved. According to the statistics of research firm IDC in the third quarter, since the addition of Arm-based M-series chips to the Mac lineup, Apple has almost doubled its share of the PC market in three years.
The three giants of chips empathize with the X86 architecture and begin to develop Arm architecture chips
Since 2016, Microsoft has invited Qualcomm to use the Arm architecture to make chips for Windows PCs. Prior to this, Windows PC chips were all based on Intel's x86 architecture, which was also known as the "Wintel Alliance", which was once regarded as an absolute monopoly in the PC industry.
According to sources, Microsoft gave Qualcomm an exclusive license to cooperate at that time, but the agreement will expire in 2024. As the deadline approached, Microsoft began to encourage new contenders to get involved. Reuters exclusively reported that Nvidia and AMD, two chip giants, are among the invitees, and they will both develop Arm-based PC chips, which are expected to be launched in 2025.
Of course, Qualcomm, which has taken the lead, is not to be outdone. At the same time on Tuesday, Qualcomm unveiled the Snapdragon X Elite chip for PCs at the Snapdragon Summit, which is based on the Arm architecture and can process AI models with 13 billion parameters to complete AI tasks such as summarizing emails, writing text, and producing images. Laptops equipped with the chip will be available next year.
Why did the chip company design the CPU for Microsoft?
Microsoft took the chip company to bet on Arm-based chips at this point in time, of course, not only to get rid of the dependence on Intel's X86 architecture, after all, the Wintel alliance has been maintained for nearly 40 years, and the two sides have already passed the honeymoon period.
A more plausible explanation is that Microsoft is preparing for generative AI applications on the PC side. AI-enhanced software such as Copilot from Microsoft will become an increasingly important part of the Windows system. As a result, Microsoft has been encouraging chip companies to add advanced AI features to the CPUs they are designing, and upcoming chips from companies such as Nvidia and AMD will need to invest resources to make this happen.
And Apple's M-series chips are a good example. On the one hand, the M chip developed based on the Arm architecture has stronger performance and lower energy consumption, making it very competitive in mobile scenarios. On the other hand, thanks to the "Unified Memory Architecture" technology adopted by M-series chips, Macs have an advantage when running generative AI locally, which requires a lot of memory.
What are the possible challenges?
However, Microsoft's plan to invite chip companies to develop PC chips is not guaranteed to be successful. Developers have spent decades and billions of dollars writing code for Windows, which runs on the x86 computing architecture, which is used by both Intel and AMD. Computer code built for x86 chips doesn't automatically run on Arm-based designs, and the shift can be challenging.
In addition, Intel has also been experimenting with incorporating AI features into its chips, and recently it showed off a laptop that runs ChatGPT-like features directly on the device.
Reference Links
https://www.reuters.com/technology/nvidia-make-arm-based-pc-chips-major-new-challenge-intel-2023-10-23/
Qualcomm Releases Mobile Chip Supporting Generative AI
On October 24, Qualcomm announced two new flagship chips during the Snapdragon Summit, namely the Snapdragon X Elite chip for laptops and the Snapdragon 8 Gen 3 chip for mobile terminals. Both of these products released by Qualcomm are designed with the needs of generative AI in mind.
How does the Snapdragon X Elite chip perform?
Previously, Neocortex reported that the Snapdragon X-series chips will use a custom Oryon CPU core chip (based on the Arm architecture), which is mainly used on laptops running Microsoft Windows, with the goal of competing with Apple computers equipped with M-series chips. According to Qualcomm, "The Snapdragon X Elite is the fastest CPU in the PC space, faster than anything from Apple, AMD, or Intel, with the goal of pushing the limits of what a PC can do while delivering excellent battery life."
· At the same power level, the Snapdragon X Elite chip performs twice as fast as Intel's 13th Gen Core i7-1360P and i7-1355U processors, while consuming 68% less power.
· In terms of single-threaded performance (referring to the speed and efficiency of a processor when handling a single task), the Snapdragon X Elite chip beats Apple's M2 Max chip.
· In terms of peak performance (the highest level of performance a processor can achieve in an ideal state), the Snapdragon X Elite chip is on par with Apple's M2 Max chip, while consuming 30% less power.
· When it comes to generative AI, the chip can run generative AI models involving more than 13 billion parameters, and its AI processing power is 4.5 times that of other laptop chips. PC products powered by Snapdragon X Elite are expected to launch in mid-2024.
How does the Snapdragon 8 Gen 3 chip perform?
The Snapdragon 8 Gen 3 chip embodies Qualcomm's ambitions in generative AI: it wants to pull the generative AI experience from the cloud to on-premise devices. This is similar to Google's self-developed Tensor G3 processor on the latest generation of smartphones, the Pixel 8 and Pixel 8 Pro, which Google sees as a key fulcrum for bringing generative AI capabilities to mobile phones.
The Snapdragon 8 Gen 3 is based on TSMC's 4nm process. According to Qualcomm, the Snapdragon 8 Gen 3 has a 30% increase in peak CPU performance and a 20% improvement in energy efficiency compared to the previous generation.
The Snapdragon 8 Gen 3 is powered by the Adreno 750 GPU, which delivers 25% faster graphics rendering and 25% more GPU energy efficiency, as well as enhanced gaming and streaming capabilities.
The Snapdragon 8 Gen3's Neural Processing Unit (NPU) is a Hexagon NPU, which is 98% faster and 40% more energy efficient than its predecessor. The role of NPU in the field of AI is to accelerate deep learning computing, provide efficient neural network training and inference, and complete intelligent tasks such as image recognition and speech processing.
Thanks to powerful GPU and NPU support, the Snapdragon 8 Gen 3 can run generative AI on the device side, and it supports models with more than 10 billion parameters, such as Meta Llama 2, which supports 20 tokens per second. Qualcomm executives demonstrated the Snapdragon 8 Gen3's extreme speed for generative AI tasks — using Stable Diffusion to generate images on smartphones in less than a second.
The Snapdragon 8 Gen3 is also equipped with a leading Image Signal Processor (ISP) that generates AI-enhanced photos and videos in real-time. It also allows image and video editing to be done through voice.
Qualcomm said mobile devices with the chip will be available in the coming weeks, with brands and manufacturers using it including Sony, Asus, OnePlus, OPPO, Vivo, Xiaomi, Honor and ZTE. Xiaomi President Lu Weibing was invited to participate in the Qualcomm Snapdragon Summit, he officially announced that the Xiaomi 14 series will be equipped with Snapdragon 8 Gen 3, and showed off the Xiaomi 14 real phone on stage.
Reference Links
https://www.qualcomm.com/products/mobile/snapdragon/smartphones/snapdragon-8-series-mobile-platforms/snapdragon-8-gen-3-mobile-platform
U.S. AI chip export controls came into effect on October 23
On October 24, chip company Nvidia said in a filing that U.S. export restrictions on the sale of its high-end AI chips to China had come into effect on Monday as regulators brought forward the deadline. The restrictions were set to take effect 30 days later, starting Oct. 17, when the Biden administration renewed its ban on importing advanced AI chips designed by companies like Nvidia from countries such as China, Iran and Russia. The new restrictions will prohibit Nvidia from exporting modified advanced AI chips, the A800 and H800, both of which were designed specifically for the Chinese market to comply with previous restrictions. Nvidia's A100, H100 and L40S chips are also affected by the limitations.
Nvidia expects the early ban to have no impact on its bottom line, but it did not say why the U.S. government is speeding up the process. Chip companies AMD's MI250 and upcoming MI300 chips, Intel's Gaudi2 chips have also been affected by the ban. The Commerce Department declined to comment.
-END-
We are a new content IP born in the GPT wave and incubated by YiMagazine.
Like every reader who cares about technology and the fate of mankind, we hope to better understand the rapidly changing world of technology and ourselves as "advanced intelligences" in this era of uncertainty.
Under this goal, we plan to report and discuss issues related to "intelligence" from multiple perspectives such as academic, business, ethical, and regulatory. Note that when we say intelligence, we don't just talk about AI.
To communicate with reporters, you can add WeChat (please note the company name and name):
Jeff Wang's WeChat: wjfsty
Zhang Siyu WeChat: helianthus351
Wu Yangyang WeChat: qitianjiuye
The copyright of this article belongs to Yicai.
It may not be reproduced or translated without permission.
You can purchase the October 2023 issue of CBN magazine