当前位置: 主页 > 成功案例 >








不久的将来 芯片级AI将成主流手机厂商标配

作者:admin  日期:2018-12-07 16:06  人气:
The controversy began when AI mobile phones first came into people's sight.
2016年底,华为荣耀发布了荣耀magic AI手机,成为业内第一个吃螃蟹的人,但因复杂原因销量不佳。外界不乏质疑,即使在华为内部,团队也承受了不少压力。
At the end of 2016, Huawei Glory released its glorious magic AI mobile phone, becoming the first person in the industry to eat crabs, but sales were poor for complex reasons. There is no lack of doubt from the outside world. Even inside Huawei, the team is under a lot of pressure.
"Fear that, like the dual cameras in those days, they will not get the market effect." Zhao Ming, CEO of Glory, recalled that the whole project was going very hard.
But less than a year later, the AI race quietly started at the chip level. In September 2017, Huawei and Apple released their respective AI chips one after another. Qualcomm and Samsung, the traditional chip giants, were unwilling to lag behind. They launched chips with AI algorithm to wait for the opportunity to catch up.
How on earth AI mobile phone is still unclear, head players have rushed to the fore. Whether AI processors are integrated in mobile chips will become the key point of differentiated competition in the industry.
Cell phone companies, constrained by chip suppliers, are also exploring. Or build your own AI engine like vivo, or a voice assistant like millet, from the broader Internet of Things.
Almost all mainstream players realize that AI broadens the capabilities of mobile terminals, just as smartphones overturn functional machines. This is not only an arithmetic revolution, but more importantly, mobile phones extend from information terminals to life service platforms, which may reshape their position in the entire information industry chain.
In May this year, a year and a half after the introduction of AI mobile phones, China Telecom released the White Paper on AI terminals, which for the first time attempted to define AI mobile phones from the perspectives of computing power, capability and application.
A mobile phone industry insider told reporters that today, even if they do not have AI, there will not be any mainstream mobile phone companies dare to say that they do not support AI. CounterPoint, a research institute, said that by 2020, a third of smartphones will have AI capabilities, with more than 500 million.
However, AI is not only the enhancement of image and voice recognition functions, but also a long way from the real landing.
Mobile AI brain
Until now, there is no uniform and clear concept and standard for AI mobile phones. But the earliest version of AI mobile phone started with the chip of AI mobile phone.
In April 2017, Apple Chip Manufacturer TSMC began to produce A11 chips, and shipments reached 50 million by July. The chip with a built-in neural network computing unit was launched into the next generation of the iPhone in September of that year.
Huawei launched the Kirin 970 chip, which encapsulates a dedicated NPU computing unit and has begun to be widely used in Huawei and its glorious mid-and high-end products.
Both sides claimed that they had released the world's first AI mobile chip, but at that time, there was no consensus on what the AI mobile phone was and what the essential difference was between the AI mobile phone and the smartphone.
2016 is the first year of AI, and hot money is pouring in. According to the third-party financing Statistics Platform whale quasi-show, the total investment in the field of AI is 122.4 billion yuan.
The mobile phone industry has always been the pioneer of new technology, but how much change AI will bring, most manufacturers have no idea.
Since the fourth quarter of 2016, the global mobile phone market has entered the downward channel, with negative growth for the first time, with shipments falling by 0.5% year-on-year. Like concepts such as "curved screen", AI was initially seen only as a way to stimulate the market. Most manufacturers either carry cloud-based AI or access third-party AI applications. What people can remember is the faster speed of mobile phones and the increased battery life.
But in more and more Internet user scenarios, data transmission and operation will be more time-consuming, and power consumption is huge. The computing power of cloud AI alone cannot meet the demand.
Artificial intelligence expert Liu Feng explained to reporters that if the Internet is regarded as a brain, cloud AI is the central nervous system, and the AI chip set in the front end is more like peripheral nerve.
If AI is regarded as the computing power attached to the mobile phone system, the chip is the core of computing, which is equivalent to the brain of the mobile phone AI.
That is to say, in terms of algorithm principle, cloud AI is not essentially different from the AI integrated on chip, but the latter has become the key to improve the computing capacity and reduce power consumption of mobile phones.
华为称,单独搭载NPU的麒麟970芯片能在每秒钟识别2000张图像,比搭载普通CPU模式的芯片快20倍。苹果的A11每秒运算次数最高可达 6000 亿次。这样的算力,云端AI无法提供。
According to Huawei, Kirin 970 chips with NPU alone can recognize 2,000 images per second, 20 times faster than chips with normal CPU mode. Apple's A11 can run up to 600 billion times per second. Cloud AI cannot provide such computing power.


上一篇:猪价回暖或带动6月CPI小幅上行 未来物价走势如何