Nowadays, AI technology has begun its productization practice from theory, and it has been widely used in the consumer electronics field. Especially in the mobile phone industry, the hottest technology application is AI, up to the flagship, and down to thousands of machines. To play the game, the entire industry is eager to grab the AI dividends, and even the AI mobile phone has become a company with technical know-how.
The purpose of this article is to clarify the true and false AI of the mobile phone industry by demonstrating the ambiguity and commotion of the current mobile phone industry.
AI is the development direction of the next generation of smart phones
AI, artificial intelligence, English Artificial Intelligence, also known as machine intelligence (Machine Intelligence). As the name implies, this is the intelligence that the machine exhibits, which is different from the natural intelligence (Natual Intelligence) of human or animal.
In the past two years, research on artificial intelligence has mushroomed, and it is precisely the unlimited prospect of AI. Each of the giants has invested heavily in AI research: Google proposed the 'AIFisrt' strategy; Apple is in its latest processor. In the integrated neural engine NE, launched the CoreML suite to run machine learning tasks, similar to Apple's neural engine; Huawei launched a Kirin 970 chip last year equipped with a powerful embedded neural network processor NPU; FaceBook and Microsoft are a few Years ago an AI research team was established.
But the key to 'artificial intelligence' really opening the door to the lives of hundreds of millions of ordinary people is the smartphone.
Since 2016, the smart phone market has seen a period of plateau, leading to the underlying cause of the entire growth stagnation is still weak technical innovation, micro-product innovation can no longer meet user needs, the entire smart phone industry encountered the ceiling.
Also at that time, Huawei, which had already laid out AI in advance, declared that the next generation of smart phones will be AI phones, and Huawei’s Internet mobile phone brand glory won the first game. In 2016, it released an exploration of the future. Craftsmanship, Magic Smartphone for AI smart phones, this phone has autonomous sensing capabilities and can actively provide services for users, including semantic recognition, face recognition, etc.
2017 is the year when artificial intelligence broke out. Artificial smart phones have really entered the mature application stage. When smartphones have fallen into a 'high reaction', Apple, Huawei, and Glory have actively explored new technologies and introduced new products, stimulating artificial labor. The overall boom of smart phones: Apple developed the A11 bionic artificial intelligence chip and applied the technology to the newly released iPhone X. Huawei released the world's first artificial intelligence processor with a neuron network unit (NPU), the Unicorn 970 chip. And used in Mate10 products; glory in 2017 released the second generation of artificial smart phones glory V10 also built Kirin 970 chip.
The breakthrough of the artificial intelligence technology from iPhoneX to Huawei Mate10, Glory V10, etc., is indicating a new path for the future development of smart phones.
However, with the mobile industry giants taking the lead in the AI field, the incompetent mobile phone manufacturers in the highly competitive mobile phone market have begun to follow the lead. Nowadays, AI becomes the standard at all kinds of new mobile phone conferences. It seems that if you don't have a little artificial intelligence, you will be embarrassed to say hello to everyone. It's a mess, a mess, it's hard for consumers to be misled.
AI chaos, no need to false to save the truth.
1. There is a real AI chip is an AI phone
A real AI mobile phone, its first premise is that the chip it is equipped with supports AI application calculations.
Because the nature of AI is computational, only the chip has a powerful AI capability, and its AI function can be achieved.
Current mobile phone chip AI computing solutions are mainly divided into two types: independent processing units and distributed heterogeneous computing. The former is represented by Huawei's Unicorn 970 and Apple's A11 Bionic, while the latter is represented by Gaotong Xiaolong 845. For example, an SOC with an independent neural network processing unit can be used to accomplish specialized tasks. This is like a school where teachers in each subject specifically teach classes in their own subject, language teachers specifically teach languages, and math teachers teach mathematics; However, there is no independent processing unit, and the implementation of distributed heterogeneous computing SOC, we must face some AI work requires CPU, GPU, DSP and other modules to work part-time, language teacher to teach mathematics part-time, efficiency and effectiveness will of course greatly reduced .
In the Unicorn 970, in addition to the CPUs, GPUs, ISPs, and other modules of traditional mobile processors, a dedicated independent hardware processing unit, NPU, is integrated. The NPU independently completes specialized AI computing tasks to comprehensively improve AI computing capabilities. NPU features are Full-time and high-efficiency—As long as it involves AI task processing, it will be handed over to the NPU independent chip for full-time operation processing. Under the same task conditions, the NPU's operation efficiency is much higher than that of CPU+GPU+DSP distributed heterogeneous computing. The official said that NPUs born for AI deal with 25 times the performance of traditional CPUs in machine learning, 6.25 times the performance of GPUs, and have the advantage of ultra-low power consumption. For example, in the calculation of image recognition, The Unicorn equipped with NPU can process 2005 photos per minute, and the CPU operation can only reach the speed of 95 sheets per minute, which indicates that the independent NPU is powerful.
Apple's A11 Bionic also has an independent processing unit, which is named 'Neural Engine' (Dual Core Engine). But to compare AI's power, the Unicorn 970 is about three times that of the A11 Bionic.
Below, we knock on the blackboard, note the key: Floating-point computing capability is one of the important manifestations of SOC calculations, and when measuring floating-point computing capabilities, there are different standards of accuracy - FP16 (half-precision floating-point) is the current embodiment of a The main indicator of SOC AI power is TFLOPS, which means tera floating point operations per second, meaning 1 trillion floating points per second. The Unicorn 970's FP16 performance reaches 1.92 TFLOP, and it can perform 1.92 trillion per second. In fact, Apple’s official said that A11 Bionic can perform 600 billion operations per second, which is only one-third of Unicorn's 970 power.
Qualcomm Xiaolong 845, a distributed heterogeneous computing solution, uses GPUs, GPUs, and DSPs to perform AI operations. It analyzes and issues specific AI tasks through a neural processing engine (SNPE) before deciding where to One or several processors are involved in heterogeneous computing. Because there is no independent AI processing module, when other tasks occupy these three parts, AI operations (such as image recognition) need to be 'queued' - or at least expressive Certain losses.
MediaTek P60 is called 'Dual Core APU' but it has no independent AI processing unit. Its original intention is to improve the efficiency of image post-processing through the ability of multiple DSPs. By the way, it can do some AI-related image processing algorithms. Its essence is two. The 'optimized' DSP overlay. The strength of DSP lies in image processing. Compared with CPU and GPU, GPU has some advantages in dealing with AI algorithm, but it is not as efficient as dedicated AI processing unit.
Therefore, in theory, Unicorn 970 is undoubtedly the best chip for the current commercial AI solution. So what is the actual running score or experience?
Unicorn 970 AI run points
Dragon 845 AI Run points
In terms of running points, we used Lu's latest 'AI run points' function to compare and found that the Kirin 970 run was divided into 233 points, while the Xiaolong MIX2s running Xiaolong 845 was run into 194 points. It is worth mentioning that Qilin 970 Compared with Xiaolong 845, which was released in the first half of the year, it is still able to lead by a large margin, which indicates its strong AI capability.
Therefore, regardless of the AI solution and implementation principle, or the actual running points and experience, the Unicorn 970 is the best AI chip. The market is not the same as the 'AI mobile phone' which does not have an independent AI processing unit. It is actually 'false'. 'AI mobile phone.
AI operating system is the AI mobile phone
AI is not only embodied in the hardware level, but also includes the support and optimization of its software. Currently on the market, only EMUI systems truly support AI capabilities. The EMUI system has features such as smart acceleration, smart services, smarter features, and AI continues to learn user habits and is able to maximize AI functionality.
EMUI supports AI engine intelligence acceleration at the system level: Based on artificial intelligence 2.0, multi-core heterogeneous resource scheduling enables application resource-aware scheduling. Based on application priorities matching resource usage requirements, it actively organizes memory fragmentation and performs background memory compression. Defragmentation mechanism of the new file system, the system scheduling performance increased by 60%, operation fluency increased by 50%, long-term use is not Caton.
Smart Service: The smarter EMUI brings ubiquitous smart services, such as automatic understanding of users' travel, courier and other daily life service information, and is presented in a negative screen to help users work more conveniently and live.
EMUI 8.0 continuously learns user preferences, understands users more and more, and brings better and better experiences: Through meticulous learning of user behaviors, intellisense and intelligent behavior prediction, EMUI's AI capabilities are unprecedented Degree of completion.
The so-called 'AI' in other mobile phone systems on the market is only limited to a single function and does not penetrate the entire system level.
For example, some mobile phone manufacturers only add similar 'AI assistant' software. They actually only help to allocate part of the system resources and do not have autonomous learning functions. The essence is the same as some mobile phone butlers just a few years ago. It is only a relatively mechanical cleaning tool.
Some other manufacturers add some so-called AI photo-enhancing functions to the camera interface. However, in reality, only the official algorithm for photo-taking is more traditional to optimize. It is not based on the AI-aware scenario to make further targeted optimization decisions.
Therefore, Huawei and Glory mobile phones equipped with EMUI system have the software foundation to support AI. With the real AI chip unicorn 970, from hardware to software to fully open up, truly artificial intelligence.
There AI mobile computing platform is AI mobile phone
AI chips or AI phones are not static, but they need to be constantly developed to tap more potential, and to graft more resources with the identity of the platform, this is the AI mobile computing platform. Its essence is to serve users, through continuously connecting more AI The developers of applications continue to innovate the application experience of AI mobile phones, bringing more intelligent user service experiences. In the image metaphor, the AI chip is essentially doing well the black soil and providing sufficient nutrients for the soil to be able to flourish. Users enjoy the value of AI openness.
In the opening of the AI mobile computing platform, Unicorn 970 is doing relatively early, and the underlying technology of AI has a relatively complete open capability. Huawei has taken the lead in introducing HiAI, an AI computing platform for mobile terminals, which enables developers to write quickly and efficiently. Artificial intelligence application running on AI mobile phone.
On March 19th this year, at the Hong Kong Developers Conference, Huawei announced HiKey970, the world's leading artificial intelligence development platform. HiKey970 is a platform that is suitable for artificial intelligence application development. It integrates Hiaro framework and other mainstream nerves of Huawei's innovative design Linaro. The network framework, in addition to supporting CPU and GPU AI operations, also supports NPU-based neural network computing hardware acceleration, enabling developers to carry out deep learning algorithms, intelligent robots and smart city development, Huawei and Glory are working to build ' The open AI ecology allows developers and consumers to enjoy more innovative applications brought about by AI scenes and experiences, making the smart life from imagination a reality.
Of course, there are Apple and Qualcomm manufacturers with software development kits: Apple has a CoreML framework and Qualcomm has an NPE neural network engine, but we have already explained that Qualcomm's SOC does not have supporting AI hardware.
So, in the Android camp, glory equipped Kirin 970 is the real AI mobile computing platform.
Summary: Based on this, in the Android camp only Huawei and Glory are emboldened to talk about what is a true AI mobile phone; Glory V10 builds a SOC based on the real AI of Unicorn 970, and Glory R&D AI was the first to start, with the EMUI of the real AI system. Its AI photography, face recognition, AI translation, AI assistant, etc. have become the most complete AI applications in the current stage, and this first-mover advantage not only makes their AI mobile phones more and more clever, but also Under the Matthew effect of computing power and data, it will speed up competitors in the future.
It is also worth noting that Huawei and Glory are more focused on the development of end-point AI. AI calculations are performed directly at the device terminals. With the help of powerful AI computing capabilities, even if there is no network support, AI operations in the future will be performed in real time. Side-by-side processing, timeliness can also be greatly improved; and without submitting data to the cloud, the user's data security is more secure.
The concept of artificial intelligence is not based on speculation, but on who can finally provide truly amazing products and services. After the first year of AI, there will be big reshuffles. More and more companies will join the ranks of AI mobile phones, but the industry’s Developments such as Ebb Tide will eliminate companies that do not possess core technology. Any progress and revolution in technology requires honest, hard-working, long-term players, those who speculate on 'AI' concepts but have little R&D investment. The next five years of the smart phone industry may soon be annihilated.