Vivo X21 used by the dragon 660 AIE is what?

Vivo released the mainstream flagship X21 in the first half of 2018 for almost a week. Before the release of this machine, I wrote an article of an outlook. From the level of my understanding, I said that X21 may have all aspects updated. From the perspective of the article, most of the articles are still on the point, but you can be proud of it. However, the adoption of the SoC, the Xiaolong 660 AIE, is the wrong one. For this issue, although it was previously issued, Long Weibo gave a brief explanation, but still feel that it is not enough to understand, simply expand it.

Re-evolution of the wisdom of both software and hardware

The SoC, which looks 'unfamiliar' and 'familiar' to the Snapdragon 660 AIE, sparked speculation on the night of the vivo X21 conference. Was it just a change of name or a new solution with a brand new AI? My long microblog says so:

With the addition of the AIE suffix, it is felt that Qualcomm is more to solve the problem of promotion of partners. Because now Apple's A-series processors, Hass and MTK all have physical AI processing units, and Qualcomm has saved costs for terminal manufacturers. , Through the HVX vector processor, CPU, GPU these computing units, plus the software SDK together. Although from the computing power and compatibility, has developed to the third generation of Nine Dragons neural processing engine is not worse than the physical unit ( The number of convolutional layers exceeds 120 levels. It was previously known that Kirin 970 is about 50 layers. The P60's MediaTek does not tell me). The compatibility is also good. After all, it has existed from the 820 era, but it still cannot solve the problem of publicity and cognition. Because consumers will say: You don't have or you don't.

So from the very beginning, the Xiaolong 660 AIE made me a little bit 'disappointed', because from the information previously disclosed, the Xiaolong 660 AIE used on the X21, and the previously introduced and recently launched models using the platform of the dragon 660 is no different, the biggest difference may be the naming method. But after further in-depth understanding, I found the hidden secret behind it: In this new era of smart phones that are not solely driven by hardware, how to use a full set of The solution enables the organic combination of CPUs, GPUs, and even HVX vector processors in the SoC, and improves the execution efficiency and results in the AI artificial intelligence applications that have been rapidly popularized. It is a common solution between processor manufacturers and handset manufacturers. Things. This is just like the same engine is assembled in different cars, may bring completely different driving and control experience. In the era of hardware and software integration, how to use deep optimization to allow SoC to maximize its characteristics to serve the consumer Experience, it is the focus.

Xiaolong 660AIE is such a "soft and hard" and experience the existence of upgrading.

At the same time as vivo X21 was released, Vivo launched a brand new artificial intelligence brand Jovi to emphasize the technical strength of vivo in artificial intelligence technology. This AI intelligence system integrates smart screen, wisdom engine, AI beauty. Behind the AI points of interest that consumers can clearly perceive are the crystallization of Qualcomm's full set of AI SDK solutions and vivo productization experience: Compared to previous models using the Opteron 660, the vivo X21 is suitable for AI applications. The experience has shown significant progress - not only faster response, but also more accurate prediction and guidance of user behavior. Of course, Qualcomm also 'enables' such a The new SoC was named: Xiao Long 660AIE. The strategic cooperation with vivo in the AI is also the first gorgeous appearance of the Xiaolong 660AIE.

Since it is a strategic cooperation between the new application model and the new strategic cooperation, how can we not have a “different” name to allow consumers to better understand the advantages of the new program? Qualcomm has eaten this loss: Remember early The mobile phone processor core battle, MT6752/6753 that Qualcomm's old rival MTK launched at that time, was once known as God U, not only the 8-core architecture allows partners to get full fire support when the terminal is promoted. The more the core, the stronger the performance. This kind of concept planted the grass in the minds of consumers - although the quad-core Xiaolong 600 didn't actually lag behind competitors in all aspects. Qualcomm had to deal with the inertial perception of market consensus. The rushed launch of the first eight-core processor, the Snapdragon 615 Series SoC, resulted in unsatisfactory performance in all aspects of hasty implementation. It was widely criticized in the industry. If it weren’t for the Snapdragon 820 that eventually relied on the quad-core architecture, the MTK was defeated. The 10-core flagship Helio X20 broke the theoretical myth that 'multi-core is strong', and then used Xiaolong 625 to stabilize its position, strongly reaps the mid-to-low end markets where MTK is best at it, and supplements MTK's shortest board with its baseband advantage. Now the market pattern of the mobile SoC is really not easy to say. Don’t you see Qualcomm’s entire mid-to-high end, including the flagship system, all back to the 8-core architecture?

And go to today's chip AI battle?? The world's major smart phone SoC genre, Apple A11 BIONIC, Huawei Unicorn 970, Samsung Exynos 9810, and even the MTK just launched the Helio P60, all have declared that they carry an independent nerve Network processor physical unit. Looking back at Qualcomm, the Qualcomm AI SDK solution started from Xiaolong 820 has reached the third generation of Xiaolong 845. No matter in terms of performance, compatibility, etc., Qualcomm's solution is in practical AI applications. Nothing happens, but as we said before, in this era of “wine is also afraid of a deep alley”, the conservative communication partner of his own family cooperated in the propaganda but they have no advantage whatsoever. And this problem, I The person in charge of the Qualcomm line was also inquired more than once, including Amon, the head of the QCT Group, and the product owner of the AI technology. The answer was still: 'We still believe that the AI SDK program is excellent. Independent neural network processor physical unit is not necessary.

Mr. Deng Xiaoping, the chief designer of the great reform and opening up, once said: 'The black cat can catch the mouse is a good cat'. Whether it is using the AI SDK to fully mobilize the CPU, GPU, HVX or even DSP's computing power The distribution achieves excellent AI computational efficiency and efficiency, or it uses the APU or NPU to perform independent calculations. As long as it allows consumers to experience completeness and excellent wisdom experience, it is a good solution. As a result, through the vivo conference, The dragon's SoC's AI capability has been pushed to a peak. Qualcomm has also played with a 'spear gun' to allow its partners to have the ability to face up against rivals. At the very least, they do not cry. Loss of money.

AI's 'Truth'

Here, I will have to parse the smart phone AI that is already everywhere, because understanding the architecture of the smart phone AI will allow us to better understand how AI is implemented on smart phones. .

A relatively simple AI application, about the composition is like this

In this structure, the well-known tens of names such as TensorFlow, Caffe, etc., are in the green part of the above figure. The computing power required for the operation needs to be down to the NPU through an API like Android NN, or Similar to the NPU's various processors to call.

The composition of the artificial intelligence engine (AI Engine) referred to in the Qualcomm Snapdragon 660 AIE is the hardware (Hexagon vector processor, Adreno GPU, Kryo CPU) + software tools (SNP SDK, Android NN, Hexagon NN), or these components, or Said that these software and hardware together, it can constitute an artificial intelligence framework that can meet or even exceed the competitor's independent NPU computing capabilities. So, why does Qualcomm insist on such an AIE architecture?

In fact, from a comparison point of view, Apple, Hass and Samsung, three smart phone manufacturers, have introduced their own NPU-equipped processor. Almost all of their own smartphone products are used. Compared with the previous CPU, GPU The biggest difference between the ISPs and the processing units is that the NPU processor functions. In addition to its inherent hardware computing capabilities, the importance of the algorithm running on it is not to be underestimated. These large companies have huge software. The team enhanced the computing efficiency of their own NPUs. For example, the number of software teams in a single Huawei accounted for tens of thousands of people. However, this terrible family had little cooperation with Qualcomm’s small partners. What do they do?

In China, there are a large number of well-known manufacturers, based on Qualcomm's AI engine, providing a wide range of solutions, such as Baidu's Duer OS for voice solutions, AliGenie for Ali, Facial++ for Face Recognition, Shangtang. Technology, camera algorithms include ArcSoft, Tencent’s security, etc. Based on the solutions provided by these vendors, various smartphone manufacturers can use their own R&D and efforts to present AI functionality in various ways. Come out, thus forming its own unique AI function and eventually forming the final product competitiveness.

The related AI functions demonstrated at the vivo X21 conference were all achieved through the mature Qualcomm Snapdragon AI engine and solutions from the solution providers mentioned above, together with the small partners of vivo's R&D department. In terms of the AI characteristics of the product itself, the efforts of Vivo's R&D part are also very important. Perhaps because of this, Qualcomm is willing to hand over such names as Xiaolong 660 AIE to vivo. At the same time, as vivo is responsible for technology research and development. The vice president Shi Yujian said: 'Qualcomm Xiaolong is a mature platform, Xiaolong AI SDK is also directly JAVA and Caffe/Caffe2 and TensorFlow these artificial intelligence framework to write applications directly, than the physical NPU closed system to open a lot. In the future, it is believed that more artificial intelligence applications will appear on the Qualcomm platform, allowing more people to experience artificial intelligence technology as the goal of vivo, and the more mature and open Qualcomm artificial intelligence engine 660 AIE is a matter of course.

See here, Dear friends should understand the truth of the so-called AI: The independent NPU is important, but how to make it with the algorithm, the solution, and finally provide the user with good AI function is the ultimate goal, at this point No matter what kind of means, as long as it can provide enough computing power, it is a good AI solution. However, personally think that Qualcomm's use of multi-core heterogeneous computing to solve the computing power of the AI engine is likely to be temporary because In the newest Snapdragon 845 platform, Qualcomm has joined a new DSP microarchitecture, achieving 8-bit performance doubling, but power consumption is twice that of the previous generation. According to this iterative upgrade rhythm, the Opteron platform will appear. Independent AI calculation units are not impossible.

Write last

Well, enter the routine BB time. Through this science article, friends should be able to have a more systematic and clear understanding of the current issue of smart phone AI, and should also be able to understand the Xiaolong 660 AIE SoC In the end it is sacred. Maybe a friend wants to say 'Isn’t such a practice changing the concept?' But I personally think that such a concept change is not just a marketing word for promotion, but Qualcomm and vivo. Can work together to provide consumers with more specific and clear information. This is by no means a mistake, but must be explained. Consumers can use this to learn about their mobile phones, from smart phones, to the 'wisdom' What is the fundamental thrust of mobile phone evolution, right?

Re-evolution of the wisdom of both software and hardware

So at first, the Xiaolong 660 AIE made me a little bit 'disappointed', because from the information previously revealed, the Xiaolong 660 AIE used on the X21, and the previously introduced and recently launched models using the platform of the dragon 660 is no different, the biggest difference may be the naming method. But after further in-depth understanding, I found the hidden secret behind it: In this new era of smart phones that are not solely driven by hardware, how to use a full set of The solution enables the organic combination of CPUs, GPUs, and even HVX vector processors in the SoC, and enhances the execution efficiency and results in AI artificial intelligence applications that are rapidly becoming commonplace. It is a common solution between processor manufacturers and handset manufacturers. Things. This is just like the same engine is assembled in different cars, may bring completely different driving and control experience. In the era of hardware and software integration, how to use deep optimization to allow SoC to maximize its characteristics to serve the consumer Experience, it is the focus.

Xiaolong 660AIE is such a "soft and hard" and experience the existence of upgrading.

At the same time as vivo X21 was released, Vivo launched a brand new artificial intelligence brand Jovi to emphasize the technical strength of vivo in artificial intelligence technology. This AI intelligence system integrates smart screen, smart engine, AI beauty. Behind the AI point of interest that consumers can clearly perceive is the crystallization of Qualcomm's full set of AI SDK solutions and vivo productization experience: Compared to previous models that use the Opteron 660, the vivo X21 is applicable to AI applications. The experience has shown significant progress - not only faster response, more accurate prediction and guidance of user behavior. Of course, Qualcomm also 'enables' such a The new SoC was named: Xiao Long 660AIE. The strategic cooperation with vivo in the AI is also the first gorgeous appearance of the Xiaolong 660AIE.

Since it is a strategic cooperation between the new application model and the new strategic cooperation, how can we not have a “different” name to allow consumers to better understand the advantages of the new program? Qualcomm has eaten this loss: Remember earlier The mobile phone processor core battle, MT6752/6753 that Qualcomm's old rival MTK launched at that time, was once known as God U, not only the 8-core architecture allows partners to get full fire support when the terminal is promoted, but also let The more the core, the stronger the performance. This kind of concept planted the grass in the minds of consumers - although the quad-core Xiaolong 600 didn't actually lag behind competitors in all aspects. Qualcomm had to deal with the inertial perception of market consensus. The rushed launch of the first eight-core processor, the Snapdragon 615 Series SoC, resulted in unsatisfactory performance in all aspects of hasty implementation. It was widely criticized in the industry. If it weren’t for the last quadruple-core Snapdragon 820, the MTK was defeated. The 10-core flagship Helio X20 broke the theoretical myth of 'multi-core is strong', and then used Xiaolong 625 to hold its ground. It was strong to harvest MTK's best middle and low-end market, supplemented by baseband advantage to storm MTK's shortest board. Now the market pattern of the mobile SoC is really not easy to say. Don’t you see Qualcomm’s entire mid-to-high end, including the flagship system, all back to the 8-core architecture?

And go to today's chip AI battle?? The world's major smart phone SoC genre, Apple A11 BIONIC, Huawei Unicorn 970, Samsung Exynos 9810, and even MTK has just launched the Helio P60, all have claimed to carry independent nerves Network processor physical unit. Looking back at Qualcomm, the Qualcomm AI SDK solution started from Xiaolong 820 has reached the third generation of Xiaolong 845. No matter in terms of performance, compatibility, etc., Qualcomm's solution is in practical AI applications. Nothing happens, but as we said before, in this era of “wine is also afraid of a deep alley”, the conservative communication partner of his own family cooperated in the propaganda but they have no advantage whatsoever. And this problem, I The person in charge of the Qualcomm line was also inquired more than once, including Amon, the head of the QCT Group, and the product owner of the AI technology. The answer was still: 'We still believe that the AI SDK program is excellent. Independent neural network processor physical unit is not necessary.

Mr. Deng Xiaoping, the chief designer of the great reform and opening up, once said: 'The black cat can catch the mouse is a good cat'. Whether it is using the AI SDK to fully mobilize the CPU, GPU, HVX or even the DSP's computing power The distribution achieves excellent AI computational efficiency and efficiency, or it uses the APU or NPU to perform independent calculations. As long as it allows consumers to experience completeness and excellent wisdom experience, it is a good solution. As a result, through the vivo conference, The dragon's SoC's AI capability has been pushed to a peak. Qualcomm has also played with a 'spear gun' to allow its partners to have the ability to face up against rivals. At the very least, they do not cry. Loss of money.

AI's 'Truth'

A relatively simple AI application, about the composition is like this

The composition of the artificial intelligence engine (AI Engine) referred to in the Qualcomm Snapdragon 660 AIE is the hardware (Hexagon vector processor, Adreno GPU, Kryo CPU) + software tools (SNP SDK, Android NN, Hexagon NN), or these components, or Said that these software and hardware together, you can constitute an artificial intelligence framework that can withstand or even beyond the competitor's independent NPU computing capabilities. So, why does Qualcomm so insist on their own AIE architecture?

In fact, from a comparison, it can be known that Apple, Hass and Samsung, the three smart phone manufacturers, have introduced their own NPU-equipped processor. Almost all their own smartphone products are used. Compared with the previous CPU, GPU The biggest difference between the ISPs and the processing units is that the NPU processor functions. In addition to its inherent hardware computing capabilities, the importance of the algorithm running on it is not to be underestimated. These large companies have huge software. The team enhanced the computing efficiency of their own NPUs. For example, the number of software teams in a single Huawei accounted for tens of thousands of people. However, this terrible family had little cooperation with Qualcomm’s small partners. What do they do?

Write last