A11 is the "artificial intelligence chip "
This time, Apple has a hardware dedicated to machine learning - the "neural engine" on its A11 Bionic chip.
Now the so-called mobile phone processors, such as Qualcomm's 835, Apple's A11, Kirin 970, etc., actually refer to a "processor package" packaged together. This computing package is professionally called Soc (System-on-a). -Chip), the statement on the tall is "computing platform"; according to the division of labor, many special function processing units are added, for example, we are most familiar with the GPU, now the number of independent units in this package has become larger and larger, such as ISP (image processing), Modem (communication module), DSP (digital signal processing), etc.
Different data comes in and is handed over to different computing modules for better results and higher energy efficiency. A11's neural engine is the same as the Kirin 970's NPU. A new hardware module that is good at neural network computing.
And this is why from 20nm, 16nm, to the current 10nm, and 7nm in research and development, major chip designers, foundry and industry are desperately trying to make the chip technology smaller, in order not to affect the chip size. Squeeze into more independent processing units.
A11's neural network engine uses a dual-core design with up to 600 billion operations per second, equivalent to 0.6TFlops (1.92TFlops for Cambrian NPU, 19200 floating-point operations per second) to help accelerate artificial intelligence. Tasks, ASICs (Application Specific Integrated Circuits/Full Custom AI Chips) specifically for Face ID, Animoji and AR applications.
With the neural network engine, Apple Senior Vice President Phil Schiller said with confidence:
“A11 Bionic is the most powerful and intelligent chip that a smartphone can have so far. ASIC-based deep learning enables high accuracy and is better than general-purpose chips (GPU, FPGA). The solution reduces power consumption."
However, Apple did not further disclose the power consumption and measured performance of this neural network engine.
A11 also supports Core ML, a new machine learning framework that Apple introduced at this year's WWDC Developer Conference, which makes it easier for developers to integrate machine learning technology into their own apps. Core ML supports all major neural networks, such as DNN, RNN, CNN, etc. Developers can package the trained machine learning model into the App.
A11 processor specificationsBoth the iPhone 8/8p and the iPhone X are powered by Apple's self-developed A11 Bionic chip. Although Apple didn't spend much time on this chip, we still know that it integrates a hardware dedicated to machine learning - "Neural Engine"
Don't underestimate this A11. Through careful research, it is not only the source of the "black technology" in iPhone X, but Apple started the technology layout nine years ago.
The parameters are unveiled.
Before introducing the "neural network engine" dedicated to machine learning in A11, let's take a look at the basic parameters of A11.
In terms of process, A11 uses TSMC's 10nm FinFET process and integrates 4.3 billion transistors (the previous generation used A10 Fusion with 16nm process to integrate 3.3 billion transistors, and Huawei Kirin 970 integrated 5.5 billion with 10nm process).
The A11 is equipped with a 64-bit ARMv8-A 6-core CPU, including two performance cores called "Monsoon" and four high-eggiciency cores called "Mistral". It is 25% faster than the previous generation A10, and the energy efficiency core is 70% faster.
Moreover, unlike the A10, A11 uses Apple's second-generation new performance controller, which allows six CPU cores to be used simultaneously, and the overall performance is 70% faster than the previous generation.
Why is it divided into performance cores and energy efficiency cores? When the mobile phone performs light tasks such as texting and browsing the web, the system will choose to call the energy-efficient core (high-eggiciency core), and when the mobile phone needs to run software with higher computing power, it needs to be used. The performance core is processed to effectively extend the average battery life. The iPhone X equipped with the A11 will be extended for 2 hours longer than the iPhone 7 when fully charged.
Another highlight of the A11 is the first time equipped with Apple's self-developed GPU, which is a 3-core GPU. The performance is 30% better than A10 Fusion, and only half of the power consumption can achieve A10 performance. This is the first self-developed GPU launched by Apple in April this year with the British GPU design company ImaginaTIon Technologies "breakup", optimized for AR, immersive 3D games, etc., 30% faster than the A10.
A11 also integrates Apple's self-developed ISP, self-developed video codec and so on. From all kinds of emphasis on "self-research" we can easily find that Apple has increasingly emphasized the autonomy of the architecture. After completely breaking up with old friend ImaginaTIon Technologies (and causing a 70% drop in share price), Apple's next self-research goal may move to baseband technology, and Qualcomm's protracted patent litigation case is a precursor.
In addition, we can also glimpse from the A11's running on Geekbench: there are several running points of A11 in Geekbench, the highest single-core performance is 4274, and the highest multi-core performance is 10438. After the value, the single core performance is 4169 and the multicore performance is 9836.
what is this concept? Compared with the previous generation A10's "single-core score 3332, multi-core score 5558", A11's performance in two aspects is close to 30% and 50% soar. The A10X single core performance in the iPad Pro averages around 3900, while the multicore performance is around 9200, still weaker than A11.
The GeekBench score of the Android camp's seed player, Qualcomm Xiaolong 835, is about 2000 for a single core and 6500 for a multicore.
Apple a11 processor detailed parametersBefore introducing this chip, let Apple use a picture tape to get a general idea of ​​this processor:
translation:
Two high performance cores (large core)
Four efficient cores (small cores)
Second generation self-developed performance controller
Self-developed GPU
Self-developed ISP (image signal processing sensor)
Self-developed video coding
Secure encryption module
A11 is Apple's first self-developed processor after cutting off the relationship with ImaginaTIon Technologies, a long-time partner. Apple's architecture and core control are all handled by Apple alone. In this regard, Apple also made several emphasises at the press conference. According to the official introduction, the A11 processor uses TSMC's most advanced 10nm process, and also uses a six-core design. Among them, the performance of large nuclear core is 25% higher than that of A10, 4 small cores are 70% higher than A10, and multi-performance processing is increased by 75%. What needs to be emphasized here is that last year's A10 just upgraded from dual-core to quad-core, and this year it jumped to six-core, and the parametric party involuntarily praised Apple. However, it seems that these two pieces of information are hidden. On the one hand, it can be seen that Apple is paying more and more attention to performance and power consumption. On the other hand, it reflects that the current application of iOS is more and more complicated and has to strengthen the core.
A11 contains 4.3 billion transistors
Having said that, the six-core design not only does not increase power consumption, but it also saves power. According to POPPUR, Apple has made a lot of effort in the tuning of the processor. It is not how to switch to the big core in daily use. In other words, except for some heavy tasks, the phone basically runs in a quad-core state. However, to explain here, Apple refers to these four "small cores" as high-efficiency cores, so you don't have to worry about performance. In addition, Apple's self-developed core architecture autonomous control design also allows for more precise control of voltage and clock. It is even possible to use ARM DynamiQ technology in advance, so that the large and small cores are also in the same cluster (cluster, file). Access is in clusters, and several sectors are combined into one cluster, which improves the efficiency of multi-core operations and reduces the delay between core switches.
The A11's GPU is Apple's self-developed three-core GPU, which has a 30% performance improvement over A10 and a 50% reduction in power consumption. In other words, A11 allows you to get longer game time while providing a smoother and more beautiful game screen.
You may also notice that this time the A11 is named A11 Bionic. Bionic is the meaning of bionics, so this processor is actually equipped with a neural network engine, and this is Apple's first attempt. The so-called neural network engine is essentially an artificial intelligence engine, but Apple is relatively low-key, and there is no propaganda everywhere. The engine primarily supports the ability to apply a Machine Learning machine learning algorithm to the TureDepth Ultra Deep Camera (for Face ID Face Recognition). In addition, during use, the engine can improve the performance of the mobile phone and the efficiency of processing tasks through machine learning such as biometrics, image recognition, photographing, and usage behavior.
Apple A11 processor running points
Although superficial, running points is still the most intuitive way to understand processor performance so far. The Geekbench database shows that the A11 processor has a single-core run of 4274 points and a multi-core run score of 10248 points. Well, this is not only a long distance away from the A10, but also directly to the 13-inch MacBook Pro. .
Apple A11 processor running points
Apple A10 processor running points
The running score of the King Dragon 835 of the Android camp is like this. . .
éªé¾™ 835 processor running points
Hf Drivers - Neodymium,Bullet Tweeter Speaker,Loud And Ultra Crisp Sound,Compression Driver
NINGBO BOILINGSOUND ELECTRONICS CO.,LTD , https://www.tweeterspeaker.com