AI computing power, new center of gravity

AI computing power, new center of gravity

But the original God also has its particularity. Better image quality is obviously a little bit, but I think the more important reason is that it spans the development background of PC, host and mobile platforms.

Spark Global limited

Although the rendering pipelines of different platforms are different, in order to ensure consistency, many 3D graphics technologies and special effects are common to developers. These technologies are nothing to put on PC and host computer, but they are completed by mobile phones, which will bring a new round of performance squeeze.


Therefore, we can see that even if the original God of the mobile terminal adopts the methods of reducing resolution, deleting a large number of modeling and special effects to ensure the basic fluency of 30 frames, the final performance is still very poor, and it is easy to overheat.


This is due to the optimization of the client itself, but on the other hand, it also proves that there is still room for improvement in the performance of mobile phone chips.

In addition, the idea of “one-time development and multi terminal deployment” adopted by Yuanshen is also a new trend in application development. All games on Apple arcade do this. If the mobile phone does not want to be the “laggard”, it is also necessary to carry some desktop graphics technology.

Application software needs a better experience, forcing hardware upgrade, which is actually the result we are happy to see. In terms of chips, this year, both Huawei Kirin 9000 and Qualcomm snapdragon 888 have chosen to improve the GPU performance. I believe it will also be a powerful impetus for the future mobile game experience.

AI, machine learning and artificial intelligence network are the fields that chip manufacturers like to mention. Both Apple A14, Qualcomm snapdragon 888 and Huawei Kirin 9000 have independent AI chips, and the performance improvement of this part has been remarkable in the past two years.

Take the neural network engine of Apple A14 as an example. Its core number has increased to 16 from 8 of the previous generation, and the peak computing power has doubled to 11 tops. The AI computing power of snapdragon 888 has reached 26tops, and the per watt characteristic can be increased by three times compared with the previous generation platform.The improvement of this multiple level is very similar to the great leap forward of CPU / GPU performance when the smart phone was just born a few years ago. As a new cutting-edge technology, artificial intelligence also begins to enjoy the dividend of this rapid progress.

However, most users don’t understand what the extremely high computing power represents and what changes will happen to our mobile phones. It is difficult for mobile phone manufacturers to provide some scenarios, which has always been a problem before.Apple has actually given a lot of popular examples on A14 this year. Take iPhone as an example, image synthesis algorithm, photo album automatic classification, power management system, and many back-end processing that we are not aware of are actually involved in machine learning technology.

To be more specific, mobile phone photography is a part that relies more on AI and we also have a stronger perception. For example, Google often says “computational photography”, Apple’s “deep fusion” synthesis technology, and Huawei’s NPU core for video rendering are all achieved by the cooperation of AI chip and ISP image sensor, which are already very mature applications.

From the perspective of human-computer interaction, some applications will use gesture tracking, semantic recognition, etc., and they will also rely on the computing power provided by AI chips. Especially in the efficiency of running specific AI algorithms, a dedicated AI core is much more effective than calling CPU and GPU directly.


5g battery life, it’s time to solve


We pursue the progress of process technology, and to some extent, we are achieving a new power balance. 5g, high screen, these two