The laptop just released by Microsoft exposes the threshold of "AI PC"

Not long ago, we put forward a point of view when analyzing the architectural ideas, product parameters, and functional design of Intel’s first-generation Core Ultra. We believe that the AI ​​acceleration function promoted by "Core Ultra" does not only rely on its built-in NPU, but uses the DLBoost instruction set of the CPU itself and the vector engine built into the GPU, and combines these three computing units (CPU , GPU, NPU) for "heterogeneous mashup"

From a positive perspective, this will enable Core Ultra to handle more types of AI floating-point acceleration codes, improving the flexibility and compatibility of its AI acceleration. But from a negative perspective, this seems to mean that Core Ultra’s AI acceleration capabilities are not exclusive, but can be shared with many previous generations of old architecture products

In fact, this is not surprising, because Intel’s CPUs have built-in instruction sets for machine learning and AI inference acceleration as early as the 10th generation Core-X series, or the 10th generation Core mobile version. Starting from the eleventh generation Core, its integrated core display has also been replaced by the Xe architecture, which theoretically has GPU AI acceleration capabilities.

We felt a little uneasy when making this inference at the time, because we only knew about the NPU integrated in the first-generation Core Ultra through some leaked images and technical information, as well as Intel’s previously launched independent AI “compute stick” Similar product names were adopted. However, this does not mean that independent AI computing sticks or accelerator cards purchased by consumers will be supported by Intel's new driver and Microsoft's next-generation AI-enabled new version of Windows system, and be able to exert their general AI acceleration capabilities

However, a recent piece of news has convinced us that we were absolutely right.

Why do you say that? Because on the evening of September 21, 2023, Microsoft officially released their latest generation workstation laptop, Surface Laptop Studio 2, and the configuration of this model can be said to have completely confirmed our previous conjectures.

As we all know, Intel’s first-generation Core Ultra is still in the early architecture disclosure stage, and the actual product has not yet been launched. Therefore, Surface Laptop Studio 2 can only be equipped with a 13th generation Core i7 mobile processor at most, and has four optional graphics card specifications: Xe core display and Geforce RTX 4050/4060/2000 Ada independent display. Rewritten content: As we all know, Intel’s first-generation Core Ultra is still in the early architecture disclosure stage and has not yet been launched. Therefore, Surface Laptop Studio 2 can only choose the 13th generation Core i7 mobile processor at most, and has four optional graphics card specifications: Xe core display and Geforce RTX 4050/4060/2000 Ada independent display

In the new version of Windows, NPU will be recognized as an independent computing unit

But what really attracts our attention is that the entire series is equipped with Intel Neural Processing Unit NPU, which is used as part of the system’s built-in AI computing power to handle functions such as camera portrait blur and automatic centering.

According to public information, the model of this "independent NPU" is Intel Movidius 3700VC, which is based on the 12nm process, and the AI ​​unit runs at 700MHz. It is worth noting that it is not much different from the previous generation Movidius MA2485 NPU, which is a 16nm process and has the same operating frequency of 700MHz. Of course, the computing power of 3700VC reaches 7TOPS, which is almost 10 times higher than MA2485, but it is not worth mentioning in front of independent graphics cards with TensorCore (taking RTX4060 as an example, its AI computing power is as high as 145TOPs).

Because of this, it can be seen from the product introduction page of Surface Laptop Studio 2 that, at least in the current version of Windows system, the independent NPU is not used as the main AI acceleration unit. Instead, it is NVIDIA’s independent graphics card and 13 The generation Core processor itself is introduced as a more important AI accelerated computing power.

So what does this show? Simply put, this seems to mean that on the one hand, future Windows systems will indeed add more AI functions, but it will most likely not require deliberate hardware upgrades. As long as you have an 11th generation or later Core processor (preferably a model with core graphics) and an NVIDIA graphics card that supports ray tracing (and TensorCore), you will most likely be able to enjoy most productivity tools driven by AI. acceleration function.

Of course, if you want to experience the feeling of using "NPU" on a Windows PC, you don't have to use a new processor. Because you can consider buying a separate accelerator card or AI computing stick equipped with Movidius 3700VC independent NPU, which will be much cheaper. Even like us, it may be possible to directly use the AI ​​computing stick integrated with the previous generation Movidius MA2485 NPU. Because it can be recognized by Windows 11 now, and it will most likely be compatible with new AI drivers and new applications from Intel and Microsoft in the future.

Please note: The pictures in this article come from the Internet

The above is the detailed content of The laptop just released by Microsoft exposes the threshold of "AI PC". For more information, please follow other related articles on the PHP Chinese website!