Is the picture good-looking? The computer drew it by itself! The first experience of AI PC

With the release of Intel's first-generation Core Ultra processors, the Core i-series CPUs we know will be a thing of the past. Since 2010, up to now, the Core i series has released a total of 14 generations (desktop), and it is not certain whether the next generation of desktop products will adopt the same naming method as the notebook side, at least from today, in the future, people will only see the chip product called "Core Ultra" in the new notebook products.

From the Core i series to the Core Ultra, the biggest change is the modular design, through the Foveros 3D packaging technology to package the SoC module, Intel 4 computing module, GPU unit and IO module together, with the additional NPU unit, the computing power of the entire processor reaches 34TOPS, providing a good hardware foundation for AI computing.

From December 15th, the world's first batch of AI PCs began to be listed one after another, and the brand will cover Lenovo, Acer, Asus, Dell, HP, Lenovo, MSI, etc., a total of 230 different models of AI PCs, among them, Lenovo's products have been listed in various online and offline channels, and the Titanium Media App has also had a hands-on experience of AI PCs for the first time.

Where do AI capabilities come from?

When it comes to the application of large model AI, everyone is most familiar with Wensheng diagrams, natural dialogues, and code generation, among which the most amazing features are Wensheng diagrams and diagrams related to image content creation.

When browsing Douyin, Kuaishou, and Xiaohongshu, you can often find a lot of beautiful pictures created by AI, many of which are based on real scenes or elements. If you don't zoom in little by little to "find faults", many people will confuse it with pictures taken or created by real people.

Specifically, the process of AI drawing includes steps such as preprocessing of input images, feature extraction, training of generator networks, and post-processing of output images.

During training, the generator network constantly adjusts the parameters to minimize the difference between the generated image and the real image, resulting in a high-quality painting effect.

In AI drawing, a computer learns how to generate a similar work by converting a painting into digitized data and then training that data using deep learning Xi algorithms. During the training process, the computer continuously optimizes the algorithm model in order to produce more realistic paintings.

In other words, the AI powered by the large model is like a student who understands the world, slowly learning, refining, and Xi the skills and styles of human painting, so as to generate artistic images.

In addition, AI can also generate new works Xi of art by imitating existing art styles, such as asking it to learn all Van Gogh's paintings, AI will convert the typical style into data, at this time, just give it a picture or photo, AI can generate a new painting with the style of Van Gogh's works.

So how fast can AI paint it?

Titanium Media App used a WeChat notebook equipped with the latest Intel Core Ultra processor for testing, here selected a model suitable for character and landscape painting, after entering Chinese keywords, it only takes less than 1 second to generate 4 pictures, and the above operations are based on local operations, without the need for networking or cloud computing power.

After understanding the principle of AI drawing and the actual experience at this stage, we can know why AI PC is warmly sought after from upstream chip manufacturers to terminal OEM brands.

First of all, with the release of the first generation of Intel Core Ultra processors, especially the NPU as a new module began to become the standard configuration of PC processors, the processing power and hardware threshold for AI tasks is rapidly decreasing.

You know,At present, the most mainstream AI painting softwareStable diffusion has no low requirements for hardware,Running memory is recommended to be more than 16GB,GPU video memory is recommended to be more than 8GB,And only NVIDIA graphics cards are recommended。 After actual testing, the notebook must be equipped with at least a RTX4060-level graphics card, so that AI can generate smaller pictures more efficiently.

This hard indicator alone directly raises the price threshold of the product to 8000~10000 yuan, and it is necessary to choose a heavier game laptop, the addition of NPU modules in the processor, and the optimization of common operations in AI computing, such as matrix multiplication, convolution, etc., so that these AI applications can achieve better energy efficiency release through a dedicated hardware computing platform.

However, despite the blessing of NPU, it does not mean that the device-side large model AI can be trained and used at any time.

Titanium Media App noticed that in the parameters officially given, Intel took the Core Ultra 7 165U as an example, its basic power consumption is 28W, but on top of this, it also marks 64W and 115W power consumption. If AI computing requires high computing power, releasing the 34TOPS computing power of the processor will put a lot of pressure on the battery life and heat dissipation performance of the notebook.

What else can an AI PC do besides drawing?

In essence, AI PC is not a "new species", as Fan Zijun, vice president of HP China Co., Ltd. and general manager of HP China's consumer products division, said at the Intel Core Ultra product launch conference: "I don't think AI PC is a revolutionary change, at most it is just one of the innovations in the development of PC technology." Because it has not fundamentally changed the application ecology of PCs and people's usage Xi. ”

So what are the differences between AI PCs and regular PCs?

From the perspective of products at this stage, it is more at the hardware layer than the application layer, that is, the PC is equipped with a series of processor products that natively support the efficient operation of AI applications on the end side, including the Intel Core Ultra series, which is called "AI PC".

As mentioned above, the hardware threshold for running AI applications was relatively high, whether it was CPU or GPU, their advantages were more reflected in data computing and graphics computing, and many architecture products were not high in the efficiency of running AI applications. It is also because of the existence of the threshold that it is difficult to popularize the application of large model AI, and many AI dialogue functions that everyone can experience today need to rely on the support of cloud computing power.

However, when the penetration of AI into daily applications and work applications gradually increases, large-scale AI that relies on cloud services will face many problems, the most obvious of which is naturally the network requirements and the inability to achieve out-of-the-box efficiency, followed by the security of various private data and the lack of customized services.

The emergence of AI PCs solves the above problems. Judging from the experience of the first batch of AI PCs, applications such as natural dialogue, code writing, image recognition, and even the creation of pictures and tracks, which used to be highly dependent on the computing power of cloud or desktop platforms, can be deployed locally on thin and light products, and the operation efficiency of the applications in the actual experience is also relatively satisfactory.

According to the introduction of relevant technicians, Intel Core Ultra can support the local deployment of large models with up to 20 billion parameters, which is completely sufficient for general personal applications.

For example, image processing methods such as masking and cutouts are often used in video editing, which used to require video editors to perform frame-by-frame operations. The local deployment of large-model AI has enabled PCs to support real-time cutout calculations. In addition, including digital humans, which have gradually become popular in the past two years, with the blessing of AI, they can also complete real-time character motion capture + virtual character generation through mobile PCs such as thin and light laptops.

The launch of more and more AI PC products will further lower the threshold for users to access large-scale model applications, coupled with the continuous expansion of built-in applications on the system side (such as Copilot in Windows 11), so that ordinary consumers can also experience AI functions and think about how to customize their AI assistants and AI productivity tools by relying on deep learning Xi.

Therefore, AI PC is just a hardware platform, which only builds a bridge between large models and ordinary users. As for whether to use it, how to use it, and when to use it, it ultimately depends on the users themselves and whether the AI applications provided by the software vendors can make users rely on them.

Michelle Johnston Holthaus, executive vice president and general manager of Intel's Client Computing Group, predicts: "AI PCs will gradually become mainstream in the next four years, accounting for 80% of the PC market share, and Intel plans to achieve AI features for more than 100 million PCs by 2025." (This article was first published on the Titanium Media App Author, Photographer/Deng Jianyun Editor/Zhong Yi)