In the iPhone 8, iPhone 8 Plus, and the upcoming iPhone X, Apple has used the latest A11 Bionic processor, which integrates a series of processing cores and complex controllers, each optimized for specific tasks. Currently, the known components include the new Apple GPU, neural engine, six-core, NVMe solid-state drive controller, and a new custom video encoder.

We need to understand these technologies one by one.

Brand New 3-Core GPU, GPGPU & ML

This is Apple’s first independently designed GPU, integrated into the A11 Bionic processor. Apple claims it is 30% faster than the Imagination GPU used in the iPhone 7, making it the most advanced graphics processing architecture in the smartphone market.

Equally impressive is that Apple’s new GPU is not only faster but also more efficient, achieving the same performance as the A10 Fusion GPU while consuming only half the power.

The GPU was originally designed to enhance graphics processing capabilities, and in recent years, it has also taken on other types of computing functions, known as “general-purpose GPU”. Apple initially used OpenCL as the API for executing GPGPU, and recently integrated GPGPU Compute into the Metal API, which is optimized for Apple iOS devices and Mac computers. The latest version, Metal 2, was introduced by Apple at the previous WWDC17 conference.

Now, Apple is designing dedicated graphics semiconductors and software to manage them, and the development speed of GPUs and GPGPUs is expected to accelerate. At the same time, Apple is also expanding its machine learning capabilities, which is one of the tasks that GPUs excel at handling. ML involves building a model based on various known entities, such as different flower photos, and using the “knowledge” model to find and identify matching targets. These targets could be flowers in other photos or images in the camera viewfinder.

Apple has not provided more technical details about the new GPU design, only stating that it has three cores. Different GPU designs are optimized for specific tasks and strategies, and defining cores in entirely different ways makes it meaningless to directly compare Apple’s GPU cores with those of Intel, NVIDIA, AMD, Qualcomm, and ARM.

Chip Intelligence Note: Although Apple has publicly indicated that its A11 integrates a GPU designed by Apple, we consulted Imagination, which declined to comment on Apple’s products but revealed that according to its agreement with Apple, Apple can customize its design based on PowerVR GPU technology. This seems to imply that Apple’s A11 GPU is still based on Imagination’s technology.

TB;DR

It is noteworthy that Apple uses Tile-Based Deferred Rendering (TBDR) on its latest A11 Bionic processor GPU Family 4 graphics processor. TBDR is a display technology created for mobile devices with limited resources. It can effectively render visible objects in a 3D scene. For example, desktop PC GPUs, Qualcomm’s Adreno, and ARM Mali execute rendering scenes in “immediate mode” for each triangle through rasterization and fragment processing stages, inputting data into device memory.

Dual-Core ISP Neural Engine

Having only a brand new GPU architecture is not enough to support the innovations of this new processor. Therefore, the A11 Bionic also includes a brand new neural engine image signal processor, which adjusts specific allocation issues, analyzes and computes thousands of image reference points, and utilizes camera sensors.

These tasks can be sent to the GPU, but the logical optimization for matrix calculations and floating-point processing allows the neural engine to excel in handling these tasks.

The neural engine itself is equipped with two parallel cores to process real-time data, capable of executing 600 billion operations per second. This means that in addition to adding complex effects to photos, it can also directly add effects to video playback, as Apple has done in previous generations of services. Besides effects, it can also enable the camera system to recognize objects and their specific components within a scene, tracking and focusing on the subject being photographed.

This neural engine is considered to represent the latter half of the A11 Bionic processor’s name, with “Bionic” evoking the idea of surpassing human capabilities, and can also be seen as a machine with human-like functions. Alternatively, the chip can be viewed as enhancing human capabilities, allowing users to accomplish tasks that ordinary machines cannot.

Brand New Six-Core CPU, 2G Performance Control

The third highlight of the A11 Bionic is Apple’s customization of ARM architecture CPU cores. Apple first launched its custom A4 processor in 2010 and began iterative updates. With the introduction of the first 64-bit A7 processor, the entire mobile chip market was thrown into chaos.

Last year’s A10 Fusion gained its name from a new architecture that managed the distribution of tasks between two performance cores and efficiency cores, achieving a balance between high efficiency and low energy consumption, allowing for more flexible performance allocation.

This year, Apple introduced the second-generation performance controller concept, which aims to extend tasks to lower-power cores or elevate workflows to faster high-power cores, even allowing all six cores to operate simultaneously. Using this asymmetric multiprocessing technology, the A11 Bionic processor can increase any number of cores based on different task loads.

In the expansion queue of multiple cores, what is needed is not just multiple cores of the processor; applications and systems must also be included in the design phase. This is where Apple’s collaboration at the operating system level with third-party developers has become a significant advantage for iPhone over the years.

Apple has detailed its software system strategy, shutting down unnecessary processing units, improving process efficiency, and ramping up as quickly as possible. Now, this functionality has also been implemented in the processor, while other mobile device manufacturers, including Samsung and LG, have generally not developed processors for systems.

Google’s Android system uses the portable JavaME platform, which has an inherent performance disadvantage. It cannot perform true tablet or desktop computing tasks but can keep the average price of smartphones below $300. The Android One platform can even be as low as $100. Compared to Android users, who are more appealing to advertisers, it does not cater to users with higher demands for user interface, application performance, and complex features.

Apple believes that the two performance optimizations of the A11 Bionic improve performance by 25% over last year’s A10 Fusion processor, and the greater benefits come from the efficiency cores, which have doubled in number, resulting in a fourfold increase and a speed boost of 70%.

In GeekBench tests, comparing the scores of the iPhone 7 and iPhone 8 (with the same RAM and resolution), the A11 Bionic processor’s single-core score increased by 25%, while the multi-core score surged by as much as 80%.

This is particularly noteworthy because Apple’s new processor also provides additional functionalities such as the neural engine, GPU, camera, and video encoding, all of which exceed half of the processor’s coverage.

In stark contrast, Samsung has been marketing the concept of “octa-core” processors for years, but in reality, core performance is slower, and the operating system has not been optimized to leverage the advantages of eight cores in practical applications beyond benchmarking. Google even emphasized at its own launch event that the Nexus 7 had sixteen cores (including CPU and GPU cores), which is purely a gimmick that does not translate to faster speeds. Moreover, over time, performance tends to degrade.

Apple’s promotional focus is on the user experience of actual applications rather than overly hyping abstract technical specifications. For example, Apple states that the A11 Bionic processor enables stunning 3D gaming and AR experiences, and that is indeed the case.

Unlike CPUs, Apple has also incorporated a secure enclave processor in the A7 processor to handle sensitive data, including fingerprint recognition, in an isolated manner.

Solid State Drive Controller

The A11 Bionic also features some other special functionalities, including an ultra-fast SSD storage controller designed with custom ECC algorithms. This design is not only for speed but also to ensure that storage media lasts longer, maintaining performance consistent with the CPU.

In other words, data stored on the device (documents, apps, and photos) can better avoid loss due to failures and wear. It’s worth noting that many Android users complain about inexplicably slowing devices over time.

Apple introduced its custom NVMe solid-state drive storage controller in 2015, capable of optimizing read and write operations from solid-state drives.

Subsequently, Apple integrated this technology into the A9 processor of the iPhone 6s. NVMe technology was initially developed for enterprise users, not consumer electronics. Therefore, at the time, there was no ready-made solution for adding NVMe controllers to smartphones, only outdated SSD storage protocols, and Apple had to manufacture and write the product code itself.

The A11 Bionic is equipped with Apple’s third-generation iOS storage controller. Interestingly, Apple did not mention this at the launch event, as there were too many new technologies to introduce, making it a natural improvement.

Custom Video Encoding

Two years ago, Apple’s A9 processor included a hardware-based HEVC decoder, allowing devices to efficiently play back high-resolution content. Last year’s A10 Fusion processor introduced an updated hardware encoder, enabling the iPhone 7 to create and save content in HEVC format.

This new feature has been implemented in iOS 11, where turning on the “High Efficiency Camera Capture” option in camera settings uses HEIF for photo compression and HEVC for video recording. The advantages of these formats are that they significantly reduce the space occupied by high-resolution photos and videos.

Apple states that a 1-minute 4K resolution video recorded in the new HEVC format is only 170MB, while the previous H.264 format occupied around 350MB, more than double the former.

Clearly, when combining all of Apple’s innovations in the A11 Bionic, it results in the most powerful performance for the new iPhone, which is also what sets Apple apart.

Editor: Chip Intelligence – Lin Zi Source: appleinsider, Tencent Digital (Bear)