Home > News content

A11 chip: invincible CPU + custom GPU + AI neural engine

via:威锋网     time:2017/9/14 15:31:53     readed:1561

September 13 morning, Apple in the autumn conference officially launched a new generation of iPhone smart phones: iPhone 8, iPhone 8 Plus and iPhone X (pronounced for iPhone 10). These three new devices are equipped with a new mobile chip, Apple called A11 bionic (not "Fusion"), and that this is the iPhone on the history of the most powerful, the most intelligent chip.

Through the launch of the A11 bionic chip details, it is not difficult to find Apple's own custom chip ambition is very large, can be described as ambitious. And with the passage of time, for the semiconductor chip custom design more and more, A11 reached an unprecedented height. Apple in the conference slides appear many times and repeatedly stressed that A11 bionic chip internal CPU, GPU, performance controller, neural network unit, ISP, etc. These are Apple's own design (Apple-designed).

So, on the Apple this "smart extraordinary" A11 chip, what are the highlights? What is the important impact on the future?

Self-designed CPU core: six cores

On the A10 chip, Apple introduced the so-called "Fusion" technology, CPU with a new four-core design, with two high-performance core and two energy-efficient core. High and low efficiency of the two cores can be based on different needs, to achieve the desired performance and energy efficiency. Where energy-efficient kernels are used to cope with dense, heavy tasks and improve processing speed to ensure performance. While the energy-efficient core is used in daily affairs, low-power operation, so as to ensure the battery life upgrade, enjoy a longer single charge life time.

The latest A11 bionic chip, Apple has taken a step forward. This time the A11 bionic CPU uses a six-core design, composed of two high-performance cores and four energy-efficient cores. Compared to A10 Fusion, two of the performance core speed increased by 25%, four energy efficiency core speed increased by 70%, performance and energy efficiency performance more outstanding. The key is that Apple has also prepared a second-generation performance controller, so you can simultaneously play the full power of six cores, performance up to 70% up to meet the multi-threaded workload.

For all six CPU cores to play the performance of this point, it is not clear how much to enhance, but at least in the benchmark test, the current leak performance is quite amazing. As can be seen from the figure below (10.2, 10.3 and 10.5 correspond to three new models), A11 bionic in Geekbench 4 performance benchmark test, the average single-core score of up to 4169, and multi-core score also has 9837 nearly 10,000 scores, Spike all mobile phone industry with ARM-based design of the mobile processor.

Compared to their own, after A10 Fusion iPhone 7 Plus in single-core mode, the score is about 3514, in the multi-core was 5970. And equipped with Apple's most powerful mobile chip A10X Fusion the latest 12.9-inch iPad Pro, single-core run points for the 3924, multi-core is 9371. If the comparison with the x86 platform 2017 MacBook Pro, this equipped with 3.1 GHz Intel Core i5 processor product single-core score of 4274. Of course, multi-core or Intel is stronger, but at least A11 to the x86 camp low-voltage processor has brought a threat.

A11 bionic CPU performance surging has no doubt, but no one phone will always keep the CPU multi-core while high-performance operation. Therefore, A11 bionic will also be in the performance of the core and energy efficiency between the core of the rational allocation of tasks, intelligent and appropriate management tasks, the lightweight tasks or background tasks to the energy-efficient core to deal with, such as text messages or browse the web, many Most of the things done to the energy-efficient kernel will be able to complete, less power consumption.

In addition, Apple also announced that A11 bionic chip has 4.3 billion transistors. This means that A11 compared to A10 using a more advanced process to create the process, no accident will be TSMC's 10-nanometer FinFET process. Because thanks to the new process, can be within the established area into more transistors or advanced features. If the A11 chip to maintain the size of 125 square millimeters, at least the transistor can be increased to 6 billion, but Apple in the same chip area to do other more work.

In any case, although from the early A4 to today A11 chip, Apple is all based on the ARM authorized instruction set customization independent CPU core, but always able to maintain the industry's strongest CPU performance advantage. So, Apple has no reason not to take the road of independent customization, and will further expand the advantages. Over the past few years, competitors mostly through the custom ARM core and build multi-core CPU and A series of chip competition, at least not fall, but this year Apple to join the multi-core competition, performance lead once again expanded, for Qualcomm, Samsung, Kirin And so how will these opponents respond, it is expected.

Apple's own design of the GPU

Apple has been using the GPU graphics processing unit from Imagination Technologies. However, Apple has long been confirmed to give up PowerVR GPU, switch to independent research and development GPU graphics chip technology. Apple's move, led directly to Imagination Technologies earlier this year, was forced to sell for sale. Although Imagination Technologies that in 2018 before Apple can not design their own GPU, but Apple's move is so unexpected.

The birth of the A11 bionic chip proves that Imagination Technologies' speculation is wrong. Apple said at the press conference, A11 bionic chip also integrates Apple's own design of the three core GPU graphics processing unit, graphics processing speed compared with the previous generation up to 30% up to the maximum.

Obviously, this year's A11 bionic chip will be able to use Apple's own design of the GPU, which is indeed a lot of people unexpected, because Imagination Technologies do not think Apple can completely avoid their own IP intellectual property, patent and confidential information, if Apple From the basis of a complete redesign of the GPU architecture without infringing IP, this is definitely a super tough challenge.

In fact, Apple is not mercilessly abandoned Imagination Technologies GPU, but over the years has been done a lot of bedding. Can be said that 2013 or even earlier, Apple will have to build a number of independent GPU team moves, a lot of GPU-related mobile chip manufacturers were Apple's acquisition, from IBM, AMD, Freescale and even Imagination Technologies and other companies high-quality graphics Engineers were robbed by Apple.

When a GPU team is completed, coupled with the long-term experience of self-custom CPU, GPU related to the work of its own work will naturally start. So, in fact, starting from A8 Apple based on Imagination Technologies IP custom GPU, known as the GPU "heart" shader core and drivers are designed to complete their own. But to the A11 bionic this generation of chips, Apple really announced a completely independent design of the GPU.

Apple did not disclose too much self-designed GPU details, but said A11 bionic GPU graphics processing unit has been optimized for machine learning techniques, novel immersive 3D games and AR enhanced reality experiences. Which in the AR side, Apple said, A11 can play a six-core CPU for global tracking, scene recognition, GPU graphics processing unit can also be used with amazing 60 fps high-speed processing, while self-customized ISP image signal processing unit can also be real-time light prediction , And finally with the new double-camera and recalibrate the sensor, so that AR experience more imaginative smooth experience, and the realism reached an unprecedented height.

A11 bionic is only Apple's first attempt to design their own GPU, although it seems just started, but has played a major role in the AR, the future is bound to move forward in a more important direction. After all, Apple in order to maintain the industry differentiation, multi-year began to deploy independent basic technology, around the processor or sensor technology to work, the purpose is to master higher control, to create their own ideal graphics performance, innovative visual and graphical environment The

Neural network engine

May Bloomberg broke the news Apple has begun to test the internal neural network engine, even in the new A series of devices equipped with the test, when a lot of speculation that A11 is likely to integrate neural network engine. Facts have proved that A11 bionic to become Apple's first neural network engine SoC mobile chip. So what about this neural network engine?

Apple explained at the press conference, there is a AI artificial intelligence called machine learning, that is, let the computer through the observation of the way to learn. The neural network engine is designed for machine learning and development of hardware, it can not only perform the neural network required high-speed computing, but also has outstanding energy efficiency. In short, through the neural network engine can take the task of CPU and GPU, greatly enhance the efficiency of the chip, with less energy to complete more tasks.

In the A11 bionic, Apple said its own neural network engine with dual-core design, the maximum number of operations per second up to 600 billion times. However, Apple did not disclose the specific floating-point performance as Huawei did (when the NFL 970 NUP in the FP16 provides the computing performance can reach 1.92 TFLOPs). Apple said that its own AI unit is mainly used for machine learning tasks, to identify people, places and objects, for the "face ID" and "rhetoric" and other innovative features to provide powerful performance.

Which for the iPhone X "face ID", face ID function will project more than 30,000 non-visible infrared spot, and then get the infrared image and dot pattern transmitted to the neural network, create a facial mathematical model, and then These data are sent to the security compartment to confirm that the data is matched. Moreover, even if the appearance changes with time, it can also be adjusted to adapt.

The neural network engine is Apple's first attempt on the A series of chips, although the scope of the current service is still very limited, more designed for the iPhone X, but the future of this AI unit will soon be extended to more areas of Apple's ecological , Including unmanned automotive systems, AR enhance the reality of technology, Apple TV and HomePod speakers, so that more original blunt equipment can also be used in the same way as human interaction, become really think.

A11 so that Apple has too much "extraordinary" at the static and other opponents

In less than a decade, Apple has built a world-class chip team, and in the mobile industry has made great achievements, it is incredible. Although most of the time we only see the surface of the performance benefits, but as Apple more and more attention to the chip and internal components of custom design and research and development, in addition to unparalleled performance advantages, the specific areas of functional features Forming a huge competitive advantage.

With the best application of the ecological, the best custom-level mobile processor, the most acclaimed fully control of the operating system, it can be said that Apple's hardware and software integration capabilities more and more powerful. Many of the OME handset manufacturers are struggling to compete with it, partly because most of the smartphone makers started too late, and many vendors could not afford the relevant and huge R & D spending.

Therefore, for Apple's competitors in the mobile phone industry, directly from the Qualcomm chip company to buy the program is obviously more simple and more desirable. Take AI unit, in addition to Apple and Huawei have NPU neural network unit, Qualcomm from the beginning of 2014 also open the NPU research and development, and in the latest two generations of Xiao Long 8xx chip are reflected, such as Xiao Long 835 on the integration of the "Xiao Long nerve processing engine software framework" to provide support for custom neural network layer, OEM manufacturers and software developers can build their own neural network unit.

In fact, ARM itself has long been in the neural network on the force, this year's release of the Cortex-A75 and Cortex-A55 are integrated into their own AI neural network DynamIQ technology. ARM claims that by DynamIQ technology in the next 3-5 years to achieve 50 times higher than the current equipment, artificial intelligence performance, the specific hardware accelerator can increase the speed of 10 times. Obviously, AI, neural network unit, machine learning and so on these words, today is just the beginning.

So Apple rivals to achieve overtaking is possible to do, as Huawei bought the Cambrian NPU license, millet surging can also buy, Samsung's own custom Exynos can be based on DynamIQ technology, other manufacturers as long as the purchase of high-pass chip The Of course, Apple will not stop, and as mentioned above, Apple's huge closed-loop eco-chain is its "extraordinary" place. More importantly, Apple is gradually have more core technology, so we see the independent S series chips, T series chips and independent storage controller, the future will be more, continue to dominate.

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments