Unicorn 970 as Huawei's first mobile artificial intelligence chip, from Yu Chengdong revealed that Huawei in the development of artificial intelligence processor began, it has been much industry attention, not only because of the hot artificial intelligence concept, more importantly, the application of artificial intelligence chips in It's the first time the industry has explored mobile handsets, after all, whoever holds the lead in artificial intelligence chips is likely to get the business lead in the next generation of mobile terminals and to lead smartphones from parametric competition to more possible forms Breakthrough.
Artificial Intelligence Chip: Cloud VS Mobile
Artificial intelligence technology has developed rapidly, the technology giant has long been enacted.Artificial intelligence chips There are two scenarios, one is for high-power cloud server computing power of high-capacity chips.Other is applied to the terminal (mobile phones, tablets and other mobile devices ) Artificial intelligence chip.
Google launched the TPU for artificial intelligence, NVIDIA introduced the GPU-based artificial intelligence chips, and Intel, Microsoft and other advances in the development of cloud AI chips.However, the chip on the mobile terminal with special emphasis on low-power, high energy efficiency In contrast, AI applications require large operations and digital processing capabilities, the distance between the two so difficult to commercial success in this area example.
This is why uninfocusing was unveiled in Kirin 970. The development and commercialization of Kirin 970 will pull AI from the cloud chip to the mobile and will mean that artificial intelligence applications will reach out to a wider range of people for humankind No delay in sensory experience and decision-making services.
Mobile terminal AI chip is difficult to do? Between dimensions, size, performance, power consumption, cooling, one can not be less.
Behind Kepler's Google AlphaGo, behind the Google second generation of TPU, build machine learning supercomputer in Google cloud platform.We may compare the realization conditions of Google TPU chip and mobile phone chip, look at the Huawei 's unbelievable difficulty coefficient.
We also see that, because Google TPU chip requires a large number of complex operations, but can rely on a large area of the chip to achieve this computing power.In contrast the smart phone, subject to the area, the mentioned factors, not very good scalability The question before Huawei Unicorn 970 is how to achieve high performance between dimensions, with support for a lot of AI computing power, which will also be the highlight of IFA on September 2.
Cooling and battery life, the same need to overcome the problem.Google TPU TPU way is to use the TPU board inserted in the data center cabinet hard drive slot to use, this approach not only to meet the TPU chip power supply problems , But also does not involve high-speed computing in the heat problem.
We know that the protection of the processor speed of the two major restrictions is the heating and logic gate delay, only to protect the chip cooling problems, to better play its performance.Google TPU chip fin can be done through the heat sink, but the mobile terminal chip Said that Huawei Unicorn 970 chip between the inch of both performance, power consumption and cooling issues, but also need to consider the issue of mobile phone life, this is definitely a big challenge.
Most importantly, the experience of artificial intelligence. Google TPU service is more for the cloud, which means that the use of the scene or need to rely on the cloud, the communication network to achieve .For example, in Google AlphaGo and Li Shishi and after the Kejie Go Duel, Google TPU chip through the neural network for machine learning to accelerate, so that AlphaGo more powerful, and the entire operation of the scene or rely on communication network to complete.Compared to, Huawei Kirin 970 rumor will introduce hardware AI chip, which means A large number of operations will be put on the mobile terminal to achieve, in order to avoid cloud computing dependence on the network, as well as the possible delay issues.
Mobile artificial intelligence chip can bring what new form of mobile phone?
Unicorn 970 before the market can trigger the industry hot, so that everyone is looking forward to equipped with artificial intelligence chips, the phone can achieve what amazing experience? It is reported that Kirin 970 will face recognition, image processing and voice access to unprecedented capabilities, And this ability not only enhances the richness and efficiency of mobile application experience, but also opens up the endless imagination to explore the new AI scene and play.
Simple example, such as taking pictures in the past we use a cell phone camera, and then use Meitu software to achieve 'beauty', but with the artificial intelligence chip unicorn 970, you can take a smart way to achieve 'plastic surgery' Photo effect here is not a simple beauty effect, but to achieve the effect of the United States, which is behind the adoption of unicorn 970 artificial intelligence algorithm to calculate a large number of user data analysis derived for the majority of friends like to take pictures absolutely Is irresistible temptation.
In addition, in addition to being able to achieve the 'cosmetic' photo effect, the future may take pictures of food, the phone will automatically remind the food contains calories, and nutrients contained, it is simply the gamble of the weight-loss users.Cool experience is to join After the artificial intelligence chip mobile phone can also support the emotional recognition, can read the user's emotions, then the mobile phone will be based on the user's emotions to adapt to more suitable experience.The imaginary AI application scenarios, perhaps Kirin 970 can support, of course There may be more rich use of the scene, it is also worth the wait.
Conclusion
In the one-sided competition of homogeneity, Huawei unicorn 970 release, so that the evolution of the smart phone has more possibilities.Mobile phones in the future or will be divided into two types: artificial intelligence and non-artificial intelligence.