This year, the hottest cell phone ring two elements, nothing less than a full screen and AI. If the appearance of a full screen world, then the configuration depends on the AI.
This year's hottest three AI chips, a Huawei Unicorn 970, was mounted on the Huawei Mate 10 and the glory of the V10 body, but also the top runner-up; the other is A11, times in iPhoneX and iPhone8 / 8 Plus; the last one is the Snapdragon 845 released by Qualcomm late this year, there is no real machine equipped with the market is expected to 2018 will see the actual results.
What is the AI chip?
The AI (Artificial Intelligence) chip we are talking about here is a coprocessor integrated in a mobile phone chip and it focuses on audio, image, language capabilities, human activity prediction, accelerated database search, encryption, etc. It Although integrated in the SoC, but its performance level CPU and GPU is not much, its ability depends on the level of chip manufacturers to optimize the AI module.
Luru AI performance ranking how to come?
Lu master AI performance evaluation using the more commonly used three kinds of neural network InceptionV3, ResNet34, VGG16 specific algorithm, the machine gives the picture content identification, according to the probability of output of the possible results list.Finally, by identifying the efficiency to judge Phone AI performance, and then give the test scores.
* Inception V3, ResNet34, VGG16 these three kinds of neural networks is the most recognized test program:
One ResNet (residual network) can train very deep network by using the residual module and the regular SGD, Master Lu AI test set the depth of 34 layers; And as an open source neural network model developed by Google, with Inception after the entire The width and depth of the network structure can be expanded to bring about 2-3 times the performance increase; VGG16 is highlighted as a simple and effective model structure, a more accurate valuation, and has a good generalization of other data sets .
(The picture above shows the Huawei Mate 10 single-part mobile phone test data, the rankings reflect the average share of many mobile phones)
Snapdragon 845 AI can bring a new breakthrough?
As mentioned earlier, only Huawei and Apple are currently the only AI devices that have been listed on the market. However, Qualcomm is quite looking forward to the progress made by AI at the Snapdragon 845 conference.
The Snapdragon 845 is Qualcomm's third-generation AI mobile platform that brings nearly three times the AI overall performance improvement over the Snapdragon 835. Not only that, the Snapdragon 845 also features Apple-like Face ID capabilities, Commonly known as XR, 3D face modeling and biometrics can be achieved, meaning that future Android machines can also use facial recognition to perform a series of cell phone operations.
In addition, the Snapdragon 845 for the first time supports room-scale 6 DOF (6DoF) and instant positioning and map construction, AI facial recognition provides a higher operability, the future Android mobile phone support Face ID will not exist Technical barriers.
Even though mobile phones with AI cores already on the market are not yet perfect (such as iPhoneX's grim Face ID), the AI core in 2018 will surely become a hot spot for the handset industry in general. However, although AI core is not a project to enhance the performance of mobile phones in the traditional sense, it is an ideal mobile phone that is 'intelligent'. With the rapid development of AI cores, we can see a breakthrough in mobile phones It is not impossible.
This year, the hottest cell phone ring two elements, nothing less than a full screen and AI. If the appearance of a full screen world, then the configuration depends on the AI.
This year's hottest three AI chips, a Huawei Unicorn 970, was mounted on the Huawei Mate 10 and the glory of the V10 body, but also the top runner-up; the other is A11, times in iPhoneX and iPhone8 / 8 Plus; the last one is the Snapdragon 845 released by Qualcomm late this year, there is no real machine equipped with the market is expected to 2018 will see the actual results.
What is the AI chip?
The AI (Artificial Intelligence) chip we are talking about here is a coprocessor integrated in a mobile phone chip and it focuses on audio, image, language capabilities, human activity prediction, accelerated database search, encryption, etc. It Although integrated in the SoC, but its performance level CPU and GPU is not much, its ability depends on the level of chip manufacturers to optimize the AI module.
Luru AI performance ranking how to come?
Lu master AI performance evaluation using the more commonly used three kinds of neural network InceptionV3, ResNet34, VGG16 specific algorithm, the machine gives the picture content identification, according to the probability of output of the possible results list.Finally, by identifying the efficiency to judge Phone AI performance, and then give the test scores.
* Inception V3, ResNet34, VGG16 these three kinds of neural networks is the most recognized test program:
One ResNet (residual network) can train very deep network by using the residual module and the regular SGD, Master Lu AI test set the depth of 34 layers; And as an open source neural network model developed by Google, with Inception after the entire The width and depth of the network structure can be expanded to bring about 2-3 times the performance increase; VGG16 is highlighted as a simple and effective model structure, a more accurate valuation, and has a good generalization of other data sets .
(The picture above shows the Huawei Mate 10 single-part mobile phone test data, the rankings reflect the average share of many mobile phones)
Snapdragon 845 AI can bring a new breakthrough?
As mentioned earlier, only Huawei and Apple are currently the only AI devices that have been listed on the market. However, Qualcomm is quite looking forward to the progress made by AI at the Snapdragon 845 conference.
The Snapdragon 845 is Qualcomm's third-generation AI mobile platform that brings nearly three times the AI overall performance improvement over the Snapdragon 835. Not only that, the Snapdragon 845 also features Apple-like Face ID capabilities, Commonly known as XR, 3D face modeling and biometrics can be achieved, meaning that future Android machines can also use facial recognition to perform a series of cell phone operations.
In addition, the Snapdragon 845 for the first time supports room-scale 6 DOF (6DoF) and instant positioning and map construction, AI facial recognition provides a higher operability, the future Android mobile phone support Face ID will not exist Technical barriers.
Even though mobile phones with AI cores already on the market are not yet perfect (such as iPhoneX's grim Face ID), the AI core in 2018 will surely become a hot spot for the handset industry in general. However, although AI core is not a project to enhance the performance of mobile phones in the traditional sense, it is an ideal mobile phone that is 'intelligent'. With the rapid development of AI cores, we can see a breakthrough in mobile phones It is not impossible.
