As we all know, AI technology is already very common on smartphones nowadays. Regarding hardware, no matter Qualcomm, Samsung, MediaTek, or Apple’s A-series masters, AI computing power will be an important hardware indicator. For example, the latest flagship SoC, the AI computing power of Snapdragon 888+ is as high as 32 TOPs, which is almost 6-10 times higher than mainstream mid-range 5G SoCs (such as Dimensity 820 and Snapdragon 768G), which is far beyond the CPU. The poor performance of these components of GPU is also enough to show the importance of chip manufacturers to AI computing power.
In terms of software, AI-driven functions are now even more ubiquitous in smartphones. From the bottom-level performance scheduling, to the anti-jamming and picture optimization in the game, to the focus, multi-frame synthesis, night scene and portrait mode when taking pictures, it is usually based on AI. In some flagship models, AI algorithms have even been used to optimize the performance of video and audio in real time, making online videos look brighter and clearer, and audio sounds more detailed.
Of course, just because AI technology has greatly changed our experience of using smartphones, it will inevitably make people wonder. Why is AI on mobile phones so common, but I rarely see related feature promotions on computers?
First of all, computers actually have AI hardware, and both CPU and GPU
In the computers we use, do we have hardware with the ability to accelerate AI execution? The answer is yes.
For example, as early as 2018, the Intel 9th generation Core HEDT platform processor has integrated the DL Boost instruction set specifically used to accelerate AI code operations . According to the official data at the time, after adding this “AI acceleration” function, the AI operation speed of Core i9-9980XE reached more than three times that of Core i9 7980XE without DL Boost instruction set.
Of course, the i9-9980XE and later i9-10980XE are products for audiophiles. And the real universal processor will usher in the AI acceleration function for the first time, and it will be on the 10nm Icelake architecture in 2019. Starting from the 10th generation Icelake notebook processor, and later the 11th generation Tigerlake and the 11th generation Rocketlake architecture, the DL Boost instruction set has been added, which means that the current Intel platform, whether it is a server, a laptop or a desktop CPU , Are all ready to execute AI code.
Of course, for gamers, AI Intel CPU instruction set, although may not be famous, but AI NVIDIA graphics acceleration features to significantly broader human cicada . Since TITAN V in 2017, Lao Huang has added a Tensor Core core dedicated to AI code execution to his high-end graphics cards. In the current RTX30 series of graphics cards, Tensor Core has evolved to the third-generation architecture. It can not only be used to perform DLSS deep learning anti-aliasing effects in games, but also can realize automatic keying and recording in live broadcast software through some auxiliary plug-ins. Noise reduction and other functions.
In addition, in AMD’s Ryzen processor, although the CPU itself does not currently have the ability to accelerate the execution of AI code, the scheduler inside the Ryzen processor actually uses machine learning algorithms. In other words, it is using the AI model to optimize its own performance and power consumption level, which can also be regarded as “AI hardware” to a certain extent.
Computers do not lack AI software, but they are all professional
Since there are actually a lot of hardware with AI acceleration capabilities in the computer, there will naturally be software companies that will develop computer-use software with AI acceleration capabilities.
For example, what many friends may not know is that the Microsoft Defender anti-virus software integrated in the Windows operating system can use AI computing power to accelerate virus scanning. It’s just that it mainly adapts to the computing power of the integrated graphics card in the CPU, so for those CPUs that don’t integrate the graphics card, you won’t be able to enjoy this acceleration effect.
For example, in the 2021 version of Photoshop, the DL Boost instruction set of Intel’s new architecture CPU has been greatly optimized. According to the results of the previous official demonstration, on the 10th and 11th generation Core platforms, the new version of Photoshop can complete the super-resolution optimization of the picture in a few seconds-automatically processing a noisy low-resolution picture into sharp details High-definition pictures. In addition, in the video editing software Premiere Pro, the AI algorithm can now automatically realize the conversion of horizontal screen and vertical screen video.
Of course, for scientists and professionals who conduct AI model training, the AI software on the computer is something they will come into contact with every day. Whether it is molecular structure simulation, atmospheric data calculation using Tensor Core of NVIDIA graphics card, or AI model training software driven by some special “AI accelerator card” hardware, it has now become a daily work in the corresponding industry.
How to make AI more popular in computers? Windows ML may be the key
From the previous examples, it is not difficult to see that the reason why AI technology seems to be difficult for the general public to feel in computers is that, on the one hand, those hardware designed for AI and with AI acceleration capabilities are indeed mainly concentrated in high-end areas. . Whether it is a 10th or 11th generation Core CPU, or an RTX graphics card with Tensor Core, objectively speaking, it is indeed not a very popular product.
On the other hand, the attributes of AI acceleration hardware also directly lead to the software that currently adapts AI acceleration functions on computers, mostly in the field of high-end productivity, and even professional-level scientific research and development. This naturally makes most users “not strong” about the AI technology on the computer, and even creates the illusion that the computer does not use AI technology.
So, is there a way to make those not high-end computer hardware also have AI acceleration capabilities? In fact, there are. Everyone remembers the example we mentioned in the previous article that Microsoft uses the CPU integrated graphics card to accelerate virus scanning. In fact, it involves an AI acceleration platform that comes with the Windows system and is supported by most computer hardware at present- Windows ML.
As part of the integrated AI acceleration capabilities of Windows, Windows ML was deployed to many computers with the Windows 10 1903 update two years ago. Unlike Intel’s DL Boost and NVIDIA Tensor Core, which can only support a few high-end hardware, Windows ML can theoretically be compatible with all graphics cards that support DirectX 12 and their CPUs of the same generation. In other words, it can at least support Intel’s 6-series Core CPU, NVIDIA’s 900 series, and AMD’s RX200 series graphics cards.
At the functional level, Windows ML is not weak compared to those “professional” AI acceleration platforms. The AI model it uses is ONNX, which is very common in the industry. At the functional level, Windows ML can also achieve a series of functions including real-time image super-resolution, AI portrait recognition, and real-time video color enhancement. For example, the latest Windows 11 operating system takes advantage of Windows ML capabilities to realize the Auto HDR function that automatically enhances the image quality of old games and adapts to HDR displays.
Not only that, just recently the CEO of French independent game studio Midgar Studio revealed in an interview that now Xbox Series X/S actually has the ability to use Windows ML to improve game performance, change game screen styles in real time, and develop new AI immersive gameplay. And a series of abilities. It’s just that the potential of Windows ML is far from being fully explored by developers so far. This has led to many users who clearly have hardware that can adapt to Windows ML, and have also upgraded systems and drivers that support this technology. But until now So far, I still haven’t felt the result of “AI acceleration” on the computer.
In other words, computer hardware that supports AI acceleration is far more popular than mobile phones with AI acceleration, and it can also be traced back to older product models. However, we still need to wait patiently for developers to discover the “gold mine” of Windows ML, so that in the future, more users can experience the charm of AI.
Posted by:CoinYuppie，Reprinted with attribution to:https://coinyuppie.com/does-the-computer-have-the-same-ai-technology-as-the-mobile-phone-actually-but-not-enabled/
Coinyuppie is an open information publishing platform, all information provided is not related to the views and positions of coinyuppie, and does not constitute any investment and financial advice. Users are expected to carefully screen and prevent risks.