One of the most important innovations of the two-hour presentation of the iPhone X was not so much a talking turd, how many single-chip system with a new generation of 4.3 billion transistors — A11 Bionic — using such advanced technology was possible. But seriously, in an interview with reporters Mashable chief marketing Apple Phil Schiller (Phil Schiller) said that at present the main novelty of each new generation of the company’s products are chips.
A11 Bionic was the next important step in Apple’s vertical integration of products for complete control over all aspects of their devices. Senior Vice President, division of hardware technologies, Apple johnny Srouji (Johny Srouji) said: “Designing our own crystals began about ten years ago, because it is the best way to create a truly optimized with the hardware and software sides products Apple“.
He also stressed that the development of new chips from Apple takes about three years, so that A11 Bionic began to create in the time to market of the iPhone 6 and A8 chip. During this cycle, the plans may be slightly adjusted in accordance with the demands of the team under the leadership of Jonathan Ive (Jonathan Ive). But three years ago the decision was made about adding the crystal neural engine to accelerate computations in the field of artificial intelligence.
Updated CPU cores on a common chip area A11: 2 of 4 high-performance and energy-efficient
Of course, each generation of chips Apple is developed on the basis of the previous work, but some blocks are processed completely. For example, two high-performance cores CPU from the A10 Fusion has received a minor update, and the number of energy-efficient cores has been doubled (in A11 they became four), the opportunity to activate from one to all six cores simultaneously. Thanks to these optimizations and the new 10-nm process technology CPU in A11 Bionic began to consume less power than the A10, despite the fact that high-performance cores now 25 % more powerful and energy-efficient — 70 %.
As you can see, a considerable area A11 is the Bionic image processor
Another important innovation was significantly more powerful dedicated image processor that allows you to achieve a high-quality color camera, superior noise reduction with a lack of light, but also speed up various effects like Studio lighting in the new portrait mode. Thanks to this new ISP block for the first time in the smartphone market it has become possible to record video in 4K resolution at 60 fps or 1080p at 240 frames/s.
For many years Apple used in their single-chip systems, graphics, Imagination Tecnologies — recent example is the 6-core accelerator PowerVR GT7600. But in A11 the company decided to integrate the designed own unit GPU. This dual-core GPU, according to Apple’s 30% more powerful than used in A10 Fusion block from the Imagination, and for the same performance consumes twice less energy. Accelerator optimized for best job with low-level graphics API Metal 2 and, according to Apple, allows you to create a games console class.
Johnny Srouji noted that the company has 30 years adheres to the principle that in those areas where it believes that it is possible to implement innovations, it tries to create its own solution: a single-chip system, CPU, ISP, display and so on. The next step in this direction was the GPU, thanks to which Apple is now able to fully control the graphics on their iOS platforms, starting from the hardware to compilers, programming languages, libraries and operating system. All this is created to operate in a single optimal bundle.
New developed in the bowels of Apple graphics accelerator on the background of the overall chip area A11
A brand new unit to accelerate a specific task was a dual-core neural engine with an output of 600 billion operations per second. It effectively cope with the tasks of matrix multiplication and floating-point calculations and is used to accelerate specific algorithms related to machine learning, like Face ID, Animoji, augmented reality, Studio lighting for portraits and more. It is created for effective work with the AI library Apple Core ML.
Such accelerators is the relative novelty of the industry. For example, Google last year introduced a special hardware server accelerators TPU (Tensor Processor Unit) for calculations using its AI library TensorFlow. This year, she released the second generation of TPU, as well as mobile-optimized version of the library machine learning TensorFlowLite. Other companies are also moving in the same direction. For example, Facebook this technology is called Caffe2Go — it was in November last year and helped to create filters for photos and videos based on neural networks directly on the user’s device in real time.
Google probably plans to implement hardware blocks TPU for mobile devices, but Apple on this front was in front and was the first to integrate the accelerator into your chip for smartphones. Until now, most of the AI calculations are made in the cloud, but the performance of these algorithms directly on the device allows to reduce delays, does not require Internet connection and provides a higher level of privacy (after all, data never leaves the device).
A11 Bionic includes a lot of other units such as a digital signal processor for high quality processing of audio, various controllers, I / o, specialized algorithms correcting code (ECC) and other blocks that increase the safety and reliability of the device. For 10 years, the Apple team has done impressive work in the semiconductor region, having risen from 65-nm chips with 100 million transistors to 10 nm from 4.31 billion.
Source:
Комментариев нет:
Отправить комментарий