The generative AI, steaming hot, is now being revered as a model by chip giants who are struggling in the consumer market.
Mobile phones and personal computers (PCs) are two typical representatives. Especially in the PC industry, after being crowned with the glamorous title of “AI PC,” there seems to be a momentum to rewrite industry rules. The first to embrace this trend are several leading PC chip players.
PC chip giants are vying for the “first launch.” AMD claims to have introduced the world’s first x86 processor with a built-in dedicated AI engine; Intel emphasizes that it was the first to propose the “AI PC” concept; Qualcomm has also elevated its PC chip to become the main highlight of the Snapdragon Summit, showcasing its performance that competes with Apple and Intel PC chips.
At the recently concluded CES 2024, AI PCs were also a “cold-buster,” generating immense buzz. On AMD’s side, they introduced the first desktop PC processor incorporating an NPU, while Nvidia launched three gaming graphics cards, claiming to have the largest generative AI platform.
Chip leaders have already made bold statements: Nvidia CEO Jensen Huang predicts that AI PCs will replace traditional PCs in the next 10 years; AMD CEO Lisa Su believes AI PCs will redefine the computing experience in the coming years; Intel CEO Pat Gelsinger says Intel’s goal for 2024 is to provide all users with the world’s first and leading AI PC experience.
As the AI PC chip race heats up, it’s worth considering whether the increasingly popular “AI PC” concept can truly revive the long-stagnant PC market. Is it the prelude to a long-awaited revival and new growth for the old industry, or just a change in marketing tactics without substantial innovation?
How will the PC chip giants handle this surging demand for edge-side generative AI computing?
01
Revolutionizing PCs with AI: Are Chip Giants Truly Committed?
The PC chip market didn’t emerge single-handedly; its success is inseparably linked to the development of the software ecosystem.
Intel, which has been dominating the PC processor market for many years, initially broke through the competition with different architecture commercial processors by forming the “WinTel alliance” with Microsoft’s Windows operating system. This alliance established Intel’s x86 processors as the dominant force in both the PC and server chip markets.
Chips and operating systems are the soil, and the variety of internet applications that have grown on these operating systems have propelled PCs into countless homes and offices, becoming a standard in both work and home environments.
The rise of smartphones and mobile internet also relied heavily on these “three essentials.” The more prosperous the software ecosystem, the more user-friendly the downstream applications, the more end-users will be “stuck” to it. Building an ecosystem barrier has played a decisive role in the rise of the “trillion-dollar market cap club” members like Microsoft, Apple, and Nvidia.
Therefore, for AI PCs to truly replace traditional PCs, it’s not enough for chip companies to just promote aggressively; they must offer genuinely convincing “killer” features that persuade users to buy.
Nvidia understands this well, having reaped the benefits of generative AI’s computing power. They have taken on the role of evangelist, constantly showcasing tools and case studies over the past year, demonstrating to customers, partners, and end consumers how easy it is to develop with generative AI and how useful and user-friendly it can be.
Microsoft and a host of PC giants have also taken on the important task of prototyping applications for AI PCs.
Microsoft, wielding OpenAI’s capabilities, has continuously revamped search engines, office software, and operating systems with generative AI, painting a beautiful vision for workers worldwide of commanding AI to do tasks for them.
PC giants are also spinning new tales about AI PCs, enabling office laptops to smoothly perform tasks like text generation, summary extraction, image creation, code generation, intelligent image clipping, motion capture, video frame interpolation, and video conferencing avatars, even without an internet connection. This not only attracts more consumers but also helps expand Microsoft’s AI developer ecosystem.
While these stories are innovative, to truly persuade consumers to open their wallets, real capabilities are needed to support these narratives, which brings us to the forte of chip giants. After all, from the inception to the prosperity of software ecosystems, the support of underlying chip technology is indispensable.
02
PC and AI Collaboration: What’s Behind the Sudden Surge in Popularity?
As the introductory chapter of generative AI in the PC race was being written, PC chip giants have been extremely busy since the fall of 2023.
In terms of the battle scenario, Nvidia, as the “veteran of AI computing,” leads the way in nurturing the AI ecosystem. Intel, the leader in PC processors with decades of ecosystem buildup, has unmatched brand appeal. Other chip giants like AMD and Qualcomm are also actively building their “friend circles” to attract PC customers.
However, Intel is no longer the exclusive favorite among PC manufacturers. These companies are becoming more inclusive, mass-producing new products based on Intel chips while increasingly supporting AMD and Qualcomm PC chips.
In the PC processor race, Intel’s Client Computing Group has seen a consecutive decline in revenue for nine quarters year-over-year.
While a major reason is the weak PC market, Intel cannot just wait for a slow market recovery. Nvidia, often compared to Intel despite operating in different verticals, has overtaken them with its soaring data center business, even surpassing Intel and TSMC in the latest third-quarter revenue.
In contrast, Intel’s client computing business is its largest revenue pillar, consistently accounting for around 50%. Intel urgently needs the PC market to regain momentum.
Since first introducing the AI PC concept in September, Intel has moved swiftly, with a resounding slogan: launching the “AI PC Acceleration Program” in October, announcing a collaboration with over 100 independent software vendors (ISVs) to integrate more than 300 AI acceleration features, aiming to ship over 100 million AI PCs by 2025; in December, they launched the Core Ultra processor and announced support for over 230 AI PC models in 2024.
With this new narrative, PC manufacturers are excitedly showcasing new models. Unlike previous PCs that supported AI, AI PCs make generative AI features a key selling point, emphasizing an “evolution” in human-computer interaction experience and office productivity.
By creating local knowledge bases and customizing personal large models, AI PCs transform into personal assistants, understanding the human language better and executing tasks more efficiently. They can quickly find files, and images, fix bugs, and improve efficiency and experience in information organization, travel planning, content creation, writing assistance, code generation, knowledge management, visual design, gaming, security, live streaming, and video collaboration.
These functions require PC chips to run more of these AI functions directly on the device, offering faster and more timely responses than cloud-based solutions, and ensuring usability even when offline. Additionally, they come with built-in features to protect user data privacy.
Previously, the AI computing power provided by PC chips was limited, and running larger AI models locally often required a dedicated graphics card. However, dedicated graphics cards demand higher cooling and power, making it challenging to meet the demand for office laptops that are lightweight, low power, and have long battery life.
03
How the Top Five Chip Giants Plan to Devour the AI PC Market Pie
Currently, AI PC chips can be divided into two camps.
One camp consists of graphics card players, mainly Nvidia and AMD, which leverage their GPUs’ prowess in parallel computing. They have enhanced GPUs’ graphic rendering and AI processing capabilities, emphasizing that computers can support edge AI well without needing a dedicated NPU (Neural Processing Unit). Nvidia’s Senior VP Jeff Fisher stated during this year’s CES that Nvidia’s GeForce RTX is the world’s largest generative AI platform, claiming that RTX AI laptops can perform 20 to 60 times better than devices based on NPU.
AI technology, especially for the gaming market, has seen significant enhancements. Both chip giants have been competing in super-resolution technology. For instance, Nvidia has been promoting its AI-driven Deep Learning Super Sampling (DLSS) technology, while AMD released FidelityFX Super Resolution 3 to further enhance game frame rates and quality.
The other camp is the PC processor players. A common approach is to pack a dedicated AI co-processor NPU into the PC processor SoC, paired with increasingly powerful integrated graphics, to achieve stronger AI computing power. This caters to high-efficiency and low-cost demands while aiming to reduce PC users’ reliance on dedicated graphics cards.
The design approach for AI PC processors is increasingly resembling an enlarged version of a smartphone SoC, focusing on process technology, NPU, memory, and packaging, all with the ultimate goal of achieving more robust AI performance:
- In terms of process technology, Apple leads with its M3 series PC chips using TSMC’s 3nm process. Qualcomm’s Snapdragon X Elite, and AMD’s Ryzen 7040 and 8040 series mobile processors use TSMC’s 4nm process. Intel’s Core Ultra uses Intel 4 process for its compute Tile, while other Tiles use TSMC’s 5nm and 6nm. There are also rumors that Intel’s upcoming Arrow Lake processors for client markets might use Intel 20A and TSMC 3nm processes.
- In design, a heterogeneous integration of CPU+GPU+NPU has become standard. Apple’s M3, Qualcomm’s Snapdragon X Elite, Intel’s Core Ultra (code-named Meteor Lake), AMD’s Ryzen 7040 series, and 8000G series all have an NPU specifically designed for AI acceleration.
- Focusing on NPU, Apple’s M3 has a 16-core NPU with 18TOPS AI computing power; AMD’s Ryzen 7040 series’ Ryzen AI engine peaks at 10TOPS, and its 8040 series NPU offers 16TOPS; Qualcomm’s Snapdragon X Elite NPU goes up to 45TOPS.
- Considering memory, LPDDR5 and LRDDR5x have become mainstream, with larger memory capacity and higher bandwidth crucial for running generative AI models locally. Apple’s PC chips use its proprietary unified memory architecture, with the M3 Max chip featuring 128GB memory and 400GB/s bandwidth, capable of running Transformer models with billions of parameters. Qualcomm’s Snapdragon X Elite uses 136GB/s memory, supporting models with over 13 billion parameters on the device. AMD’s Ryzen 7040 series supports up to 256GB memory, while Intel’s Core Ultra can run 20 billion-parameter large language models locally.
- Advanced packaging technology is essential to fit different computing units compactly onto a single chip. Apple’s M series chips use TSMC’s advanced packaging technology in their UltraFusion architecture, and AMD has mastered “Chiplet stacking” in several of its chip products.
- Notably, Intel’s Core Ultra differs from most competitors, which integrate various computing cores into a single SoC using a single process technology. Core Ultra adopts a “separate Tile” strategy in client design, allowing compute, SoC, graphics, and IO Tiles to use different process technologies from various foundries, then packaged together using Intel’s Foveros technology. This approach offers greater flexibility, achieving higher efficiency and cost-effectiveness.
Besides hardware performance, the core competitiveness of AI PC chips also includes “soft power,” namely, synergy with the AI software ecosystem.
Currently, independent PC chip players are embracing the Microsoft Windows ecosystem, while Apple follows its path.
Although Apple lags behind Microsoft in the generative AI field, its ecosystem control should not be underestimated. Apple’s chips are designed entirely around the needs of its products, striving for the ultimate synergy in software and hardware integration and achieving optimal collaboration between its proprietary chips, AI framework, and applications.
Interestingly, Apple is the only mainstream PC chip player that hasn’t discussed optimizing generative AI features in its chips. When launching the M3 series, Apple emphasized AI functions like image processing, scene edit detection, and video editing. Previously, when giants flocked to the VR/AR field, Apple introduced the concept of “spatial computing.” Now, as the battle for large models rages on, it remains to be seen how Apple will carve its path in this arena.
04
Conclusion: Can Integrating AI Mega Models Truly Revitalize PC Demand?
Entering 2024, the AI PC race has begun, seemingly poised to compete with smartphones for the title of the primary device for generative AI. Chip giants are eagerly anticipating a fruitful year for PC manufacturers.
AI PC is undoubtedly an attractive concept, and various market research firms have provided positive forecasts: IDC expects that by 2024, over 70% of terminal devices in the Chinese market will be equipped with AI capabilities, with AI terminals accounting for 55% of the market; Canalys predicts that in 2024, PCs with AI capabilities will account for 19% of the market, with around 20 million AI PCs shipped. By 2027, 60% of computers will have AI processing capabilities, with shipments exceeding 170 million units; Boston Consulting Group forecasts that by 2028, AI PCs will make up 80% of the PC market.
In the past, the emergence of Intel microprocessors, in conjunction with Microsoft’s Windows operating system, marked the beginning of the PC revolution. However, times have changed. In the post-PC era, consumer electronic products have become increasingly diverse. With internet access, generative AI functions can be integrated into any form of terminal device. As the lines between work and life blur, the PC’s role is no longer irreplaceable.
While cloud-based AI is not omnipotent, the problem is that edge AI hasn’t yet emerged as a universal necessity. The edge AI functions currently heavily promoted by AI PCs are more of an emergency substitute in special scenarios like offline use, lack of GPU, or handling sensitive data, rather than a primary driver of productivity. For heavy AI users in graphic design, content creation, and video generation, relying on cloud services or dedicated graphics cards remains the preferred choice for efficiency. For light AI users, the experience of generative AI on smartphones is also improving significantly.
AI PCs bring some innovation, but are these novel features enough to sway consumers and drive a long-awaited wave of PC upgrades? And how will chip giants differentiate themselves in core performance and ecosystem in the AI PC market? Time will tell as we let the “bullets fly” for a while.
Related:
