Home > News content

Apple A11 chip depth analysis: invincible CPU self research GPU AI neural engine

via:博客园     time:2017/9/15 11:32:22     readed:1251

On the morning of September 13th, apple officially launched a new generation of iPhone smartphones at its fall conference: iPhone 8, iPhone 8, Plus and iPhone X (pronounced iPhone 10).

These three new devices are equipped with a new mobile chip, apple called A11 bionic (not)

Through the conference presentation of A11 bionic chip details, it is not difficult to find Apple's own customized chip ambitions are very large, can be described as ambitious. And with the passage of time, more and more customized designs for semiconductor chips, A11 has reached an unprecedented height.

Apple in conference slides appeared repeatedly and repeatedly stressed that the A11 bionic CPU chip GPU, and the performance of neural network controller, ISP unit, etc. These are Apple's own design (Apple-designed).

Well, about the apple

苹果

Self designed CPU kernel: six core on A10 chip, Apple launched the so-called

Among them, the high energy core is used to deal with intensive heavy tasks, improving the processing speed and ensuring the performance. The high efficiency kernel is used in daily affairs, low energy consumption, so as to ensure the battery life enhancement, and enjoy longer single charge battery life.

苹果

With the latest A11 bionic chip, Apple has taken another step forward.

This time, the A11 bionic CPU uses the six core design, consisting of 2 high-performance cores and 4 high energy efficiency cores. Compared to A10 Fusion, two of the 25% core performance speeds were increased, and the speed of the four energy efficient core was increased by 70%, and the performance and energy efficiency performance was more outstanding.

The key is, Apple also prepared second generations of performance controllers, so you can simultaneously play all the power of six cores, performance improvement of up to 70%, in order to deal with multi-threaded workloads.

It's not clear how far the six CPU cores are performing, but at least on the benchmark, the current run scores are staggering.

From the chart you can see (10.2, 10.3 and 10.5 respectively corresponding to three new models), A11 bionic in Geekbench 4 performance benchmark, single core average score as high as 4169, while nuclear scored 9837 nearly ten thousand points, all mobile phone industry with seckill authorized mobile processor design based on ARM.

苹果

Compared to our own, the A10 Fusion iPhone 7 Plus previously scored about 3514 in the single core mode and 5970 in the multicore. With Apple's current mobile chip A10X Fusion's latest 12.9 inches iPad Pro, the single run score of 3924, multi-core is 9371.

If the comparison of X86 platform 2017 MacBook Pro, this is equipped with 3.1 GHz Intel Core i5 processor single core score of 4274. Of course, multicore or Intel is stronger, but at least A11 poses a threat to the x86 camp's low voltage processors.

A11 bionic CPU performance surge is no doubt, but no cell phone will always maintain CPU multi-core while high-performance operation.

Therefore, A11 will assign the task between bionic performance and energy efficiency of core core reasonable, intelligent appropriate management tasks, tasks or background tasks to lightweight high efficiency core processing, such as texting and surfing the web, many daily things mostly to high efficiency kernel can be completed, less power consumption.

In addition, Apple also announced that the A11 bionic chip has 4 billion 300 million transistors. This means that A11 compared to A10 using a more advanced process to build, without any accident, it will be TSMC 10 nanometer FinFET process.

Thanks to the new process, more transistors or advanced functions can be inserted within the given area. If the A11 chip holds 125 mm2 size, the transistor can be increased to at least 6 billion, but Apple does more work in the same chip area.

苹果

In any case, Apple's proprietary CPU core has always been able to maintain the strongest CPU performance in the industry, though from the early A4 to today's A11 chip, Apple has no choice but to customize the ARM core based on its authorized instruction set. So, Apple has no reason not to be firm, take its own customized path, and will further expand its advantages.

Over the past few years, most competitors by customizing ARM kernel and build multi core CPU and A series chip competition, at least not strive, but this year Apple added to the multi core competition, performance advantage to expand again, for Qualcomm, Samsung, etc. these rivals will be how to respond to kylin, exciting.

Before Apple's own design of GPU, Apple has been using the GPU graphics processing unit designed by Imagination Technologies.

However, Apple has long been sure to give up PowerVR GPU and switch to its own GPU graphics technology. Apple's move led directly to the forced sale of Imagination Technologies earlier this year.

Although Imagination Technologies believes that before 2018, apple can not design its own GPU, but Apple's move is so unexpected. The birth of A11's bionic chip proved that Imagination Technologies's conjecture was wrong.

Apple said at the 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, the maximum increase of up to 30%.

苹果

Obviously, this year A11 chip will be able to use Apple's own bionic design GPU, it really surprised many, because the Imagination Technologies is not completely clear that Apple's own IP intellectual property, patent and confidential information, if Apple from the basis of a complete re design of the GPU architecture without violating IP, this is definitely a a super tough challenge.

In fact, apple is not ruthless abandon Imagination Technologies GPU, but instead, over the years has done a lot of bedding.

Can be said to be in 2013 or even earlier, Apple has a number of actions to build their own GPU team, a lot of GPU mobile chip related manufacturers acquired by apple, from IBM, AMD, fly think of Carle and even Imagination Technologies, high-quality graphics engineer by apple to grab a hand.

When a GPU team is formed and completed, coupled with the long-term experience of self customized CPU, the work related to GPU's own design will go with the flow.

So, in fact, from the beginning of A8, apple is based on the Imagination Technologies IP custom GPU, known as GPU

苹果

Apple did not disclose too many details of the independent design of GPU, but A11 said the bionic GPU graphics processing unit for machine learning technology, novel immersive 3D games and AR augmented reality experience optimized.

The apple AR said, A11 can play six core CPU global tracking, scene recognition, GPU graphics processing unit but also with a 60 FPS high-speed processing is amazing, and independent custom ISP image signal processing unit can be real-time light forecast, finally add new double photo and re calibration of sensor, AR experience more unimaginable experience, and the sense of reality has reached a new height of hitherto unknown.

A11 bionic is only Apple's first attempt to design GPU independently, although it seems to have just started, but has played a major role in the AR, the future will inevitably move forward in a more important direction.

After all, apple in order to maintain the industry difference, multi-year began deploying independent basic technology, work around the processor or sensor technology to control the higher, to create their own ideal graphics performance and innovative visual and graphical environment.

The neural network engine in May Bloomberg broke the news that Apple has started testing the internal neural network engine, even in the test equipped with the new A series chip device, then a lot of speculation that A11 is very likely that integrated neural network engine.

Facts have proved that A11 bionic became Apple's first neural network engine SoC mobile chip. So, what does this neural network engine do?

Apple explained at the press conference, there is a AI artificial intelligence called machine learning, that is, let the computer through observation way of learning. The neural network engine is a hardware designed for machine learning. It not only performs the high-speed computation required by the neural network, but also has excellent energy efficiency.

In short, through the neural network engine can undertake the tasks of CPU and GPU, greatly enhance the efficiency of chip computing, less energy consumption, faster completion of more tasks.

苹果

In the A11 bionics, Apple says its own neural network engine uses dual core designs, with operations up to 600 billion times per second. However, apple did not disclose the specific performance of floating-point operations like HUAWEI (when HUAWEI Kirin 970's NUP provides computing performance of 1.92 TFLOPs under FP16).

Apple says its own AI unit is mainly used for machine learning tasks, identifying people, locations and objects

Among them, for iPhone X

苹果

The neural network engine is the first attempt in the A series chip on apple, although the service scope is limited, more iPhone X and design, but the future of the AI unit will soon expand to more apple ecological areas, including unmanned vehicle systems, AR enhanced reality technology, Apple and TV HomePod speaker, let more abrupt equipment can also be used to interact with humans the same way, become truly thinking.

A11 makes apples too much

Although most of the time we simply see performance gains on the surface, but with apple chips and pay more attention to the internal components of the custom design and development, in addition to the functional characteristics of performance advantages There is nothing comparable to this particular field, also formed a huge competitive advantage.

With the best application ecosystem, the best custom level mobile processor, the most praised, fully controlled operating system, it can be said that Apple's hardware and software integration capabilities are becoming stronger and stronger.

The rolling level many advantages of OME mobile phone manufacturers can hardly compete with the first part of the reason is that most of the smart mobile phone manufacturers started too late, and many manufacturers can not afford the huge spending on research and development and related.

苹果

Therefore, for Apple's competitors in the mobile phone industry, direct access from Qualcomm's chip companies to buy the program is obviously simpler and more desirable.

Take the AI unit, in addition to apple and HUAWEI has NPU neural network unit, Qualcomm began also discloses the development of the NPU in 2014, and are reflected in the latest two generation 8xx snapdragon chip, such as snapdragon 835 is integrated

In fact, ARM itself has long been in the neural network on the force, this year's release of Cortex-A75 and Cortex-A55 are integrated into their own AI neural network, DynamIQ technology.

ARM claims that through DynamIQ technology in the next 3-5 years to achieve 50 times higher than the current artificial intelligence performance, the hardware accelerator will be able to respond to 10 times the speed of acceleration.

Obviously, the words "AI", "neural network unit", "machine learning" are just the beginning today. So Apple rival overtaking is possible, as HUAWEI bought the Cambrian NPU authorization, millet surging can also go to buy Samsung's own custom Exynos can be based on DynamIQ technology, Qualcomm chip can be as long as the purchase of other manufacturers.

Of course, Apple will not stop, and, as mentioned above, Apple's huge closed loop ecosystem is its

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