While the world's attention is focused on Open AI and Google, Microsoft quietly dropped a bombshell.
In the early hours of May 21st, Beijing time, on the eve of the Microsoft Build developer conference, this software giant held a special event, introducing the concept of "Copilot+PCs."
In contrast to the previously showcased but conceptually vague "AI PCs" by other manufacturers, Microsoft has integrated the newly unveiled GPT-4o at the Windows system level; on the hardware side, the built-in Snapdragon X Elite chip can locally process generative AI applications without relying on cloud computing power.
Although rumors about Microsoft upgrading Copilot have been circulating, this event far exceeded expectations.
Firstly, it was surprising how quickly GPT-4o was integrated into end products. Considering the relationship between Open AI and Microsoft, it's likely that their collaboration on edge AI has long been in progress.
Secondly, in the initial lineup of "Copilot+PCs" products, all feature Snapdragon X Elite/X Plus chips, indicating Microsoft's push for manufacturers to collectively shift towards the ARM camp.
Over the past few years, Apple has made significant inroads in the PC market with its ARM-based M-series chips, and now Microsoft has chosen to counterattack using a strategy familiar to Apple.
During the event, Microsoft executives even placed a Macbook Air with an M3 chip next to the showcased products, demonstrating how Copilot+PCs significantly outperformed the latter in image generation.
Comparison of Image Generation Speeds, Image Source: Microsoft
Microsoft's bold move in the realm of image generation speed - can Apple catch up?
The AI PC matter is finally showing some progress.
Apart from the improved processing speed of generative AI applications, what truly stands out in this presentation is Microsoft showcasing several innovative applications never seen before in the industry.
For instance, enabling your computer to "turn back time".
This feature, named "Recall", allows Copilot to assist users in finding local files, tasks, or web browsing history.
The principle behind this feature is that the computer captures and encrypts a snapshot of the screen every few seconds, storing it on the hard drive.
Imagine a scenario: a few weeks ago, you browsed a webpage about car colors, with the only memory being "saw a red car".
With such vague information, a search for "red car" could prompt Copilot to retrieve webpages from the past containing the element "red car".
Image Source: Microsoft
If there are too many results, you can further filter the desired results by selecting a specific time interval using the timeline above. It is worth mentioning that on this timeline, you can backtrack any previous tasks or even videos you have watched.
This is definitely a leap in productivity, but the downsides are also quite apparent.
Firstly, there are privacy and security concerns. If in the past, privacy and security revolved around the Cookie controversy between users and vendors, on Copilot+PCs, this data trail has evolved into "screen recording every few seconds on the user's computer."
After the event, Musk bluntly stated, "This is an episode of 'Black Mirror,' and this feature must be turned off."
Image Source: @Elon Musk, X
During the presentation, Microsoft emphasized multiple times that all recorded information will be stored locally on the computer, without any data being uploaded to the cloud. Users can also customize which types of information the system can record.
However, with all data being kept locally, will the massive amount of information stored daily end up occupying too much storage space? Microsoft did not address this question on-site, but mentioned that the "Recall" feature will officially launch later this year.
Another exciting upgrade showcased by Microsoft is the integration of GPT-4o, which finally gives us a glimpse of what a voice assistant should be like.
In a live demo, testers opened the game "Minecraft" and asked Copilot how to craft equipment. Copilot instantly provided suggestions and even warned the player to "quickly move away" upon encountering a "zombie."
Image Source: Microsoft
In fact, if we view Copilot as an NPC in a game, similar AI-driven NPCs have become increasingly common in recent years.
It is important to note that in Microsoft's demonstration, Copilot is not a pre-set process but rather called upon by players in real-time. It needs to process the player's voice commands while monitoring the game screen.
This could potentially bring about a revolutionary change in the gaming industry.
Furthermore, Microsoft has made significant upgrades to its "relatively traditional" generative AI scenarios. On the Surface prototype demonstrated live, real-time captioning now supports over 40 languages, covering scenarios such as video conferences and streaming content.
Microsoft has also enhanced the diffusion algorithm for its Paint application, allowing users to outline frameworks with simple lines and have the computer generate corresponding drawings in real-time.
Image Source: Microsoft
While it may not sound groundbreaking anymore, it's worth noting that all these features are powered by local computing power, with the Snapdragon X Elite, previously highly anticipated by the industry, supporting Copilot+PCs.
The PC industry is truly undergoing a transformation.
Two months ago, Intel held a developer conference in Taipei and, in a rare move, jointly defined the AI PC with Microsoft: it should be equipped with an NPU, CPU, and GPU, support Microsoft's Copilot, and have a physical key for Copilot directly on the keyboard.
The original Windows key has now been replaced by the Copilot key.
Through this announcement, two key pieces of information can be gleaned: firstly, in terms of defining AI PCs, Intel and Microsoft's interests are aligned, indicating that the Wintel alliance will continue to exist in the era of AI PCs; secondly, the use of heterogeneous computing platforms as the hardware foundation for AI PCs has become a consensus among various companies.
However, the challenge lies in the fact that Intel is not the only player adept at leveraging heterogeneous computing. Therefore, at today's press conference, we witnessed the "Windows + Qualcomm" combination.
Based on the information currently disclosed by Microsoft, the first batch of "Copilot + PCs" products will debut in mid-June, with major manufacturers such as Lenovo, Asus, Dell, HP, and other mainstream brands responding positively. These products will all be equipped with Qualcomm Snapdragon X Elite/X Plus chips.
Image Source: Qualcomm
This chip, released in October last year, is built on TSMC's 4nm process and features a new custom Oryon CPU with a 4+4+4 tri-cluster design, where all 12 core frequencies can reach up to 3.4GHz. It comes with an NPU capable of 45TOPS of computing power and an integrated Adreno GPU providing 4.6TOPS of power.
In comparison, the total computing power of Intel's Core Ultra commonly found in AI PCs is around 34TOPS.
Apart from its leading computing power, the ARM-based X Elite also holds a natural advantage in energy efficiency compared to various X86 architecture chips.
In fact, as early as 2012, Microsoft made a transition from X86 to ARM and introduced ARM-based Surface chips that year. However, the ARM architecture at that time used a 32-bit instruction set, leading to almost no program compatibility.
With ARM's updated V9.2 architecture last year, including the Cortex-X4, A720, and A520 processors, all designs shifted to a 64-bit instruction set. Meanwhile, major applications like Photoshop, Zoom, and Chrome have released native versions supporting ARM 64-bit.
It can be said that the transition of ARM chips from mobile to PC is an unstoppable trend today.
Considering the strong versatility and flexibility of X86 architecture chips, Microsoft's collaboration with Qualcomm may not aim to replace Intel but rather to challenge Apple.
Currently, at least compared to the M3 chip (the M4 chip has not been used in Macbooks yet), the combination of Snapdragon X Elite and Windows is significantly stronger than M3 and MacOS.
In terms of performance, whether in Geekbench or Cinebench, the Surface Pro with Snapdragon X Elite outperforms the M3 Macbook in almost every benchmark test.
Even Apple's proud energy efficiency performance is surpassed by the ARM-based Surface after the architecture switch.
Microsoft has stated that the 15-inch Surface Pro with Snapdragon X Elite achieves a battery life of 22 hours. In actual tests, compared to the M3 Macbook, the former saw a 20% increase in battery life while improving computing speed by 58%.
More importantly, Microsoft has demonstrated the real-world application scenarios and development potential of AI PCs at this event, while Apple's actions in AI still mainly focus on auxiliary functions, which may not be convincing enough for users to make the switch.
Whether it's the Microsoft-Qualcomm partnership or the active response from various manufacturers, it's evident that a tighter ecosystem alliance is taking shape within the Windows camp.
The upcoming WWDC in a month may be Apple's final chance to respond.
