This article is based on an article from the Japanese edition of Engadget and was created using the translation tool Deepl.
I was concerned that Apple's recent announcement event might not be as exciting as I thought it would be because of the low expectations for the release of the iPhone from the beginning, but the new iPad Air, which has the functionality of the Pro at a bargain price, was unveiled and the event was more exciting than I expected.
One of the things I was personally interested in was the processor in the iPad Air.
Normally, this is the time of year when new iPhone models are introduced and the latest Apple SoC (Apple Silicon) that will be used in them is a hot topic of discussion. This is because the latest processor in the iPhone is an aspect of hinting at other Apple products with Apple Silicon.
Apple develops a SoC for the iPhone, which ships in high volume, and then uses that SoC as a foundation for other products.
However, since the iPhone wasn't announced in September this year, I was wondering if the processor that would be the starting point would be announced at this point, and if so, what kind of performance and features would it have.
Apple's sense of balance in A14 Bionic
The reason Apple develops its own SoC, Apple Silicon, is to put together the necessary building blocks for the final product it wants to build.
Typically, they use a generic SoC to design their products, but Apple's approach is different. They have a target user experience and prepare the hardware, software, and services necessary to achieve that goal. They design it from the SoC level as a necessary part of developing the hardware.
There's a limit to the number of transistors that can be used in a dedicated SoC, so if you look at how Apple allocates these transistors to achieve the features and performance they need, you'll be able to get a good idea of what they're focusing on.
For example, the A12Z Bionic used in the iPad Pro is based on the A12 Bionic, the same generation of SoC for the iPhone. The former integrates 10 billion transistors, while the latter has 6.9 billion transistors.
The difference of 3.1 billion transistors is due to an increase in the number of high-performance CPU cores from two to four and the number of GPU cores from four to eight, but there are actually many other changes.
The most significant difference, other than the number of cores, is the increase in memory access bandwidth for the doubling of high-performance cores and GPUs that share memory. In other words, the entire SoC is designed to suit the balance between required functionality and performance, and the number of cores is planned accordingly.
The A14 Bionic has 11.8 billion transistors. The previous generation, A13 Bionic, had 8.5 billion transistors, so the move from a 7nm+ generation to a 5nm manufacturing process increased the number of transistors by 3.3 billion. How that increase is allocated represents a balancing act in how Apple wants to make its devices.
Limited mention of the boldly increased Neural Engine
The A14 Bionic has a high-performance CPU core that is 40 percent faster than the A12 generation (an estimated 16.6 percent faster than the A13 generation) and a 30 percent better CPU (an estimated 9 percent better than the A13 generation). The matrix operation accelerator that accelerated the machine learning process introduced in the A13 generation has also been updated to the second generation, and the Image Signal Processor (ISP), which has significantly improved the iPhone 11's camera quality, is said to be more capable.
While these will certainly enhance the ability of iPhones to be equipped with this SoC, it may seem a bit underwhelming compared to the range of performance improvements in the past. However, Apple's high-performance cores have always been unusually powerful for smartphones.
We should think of this update as allocating incremental transistors to other elements of the device. The change in the A14 Bionic that seems to have allocated the most transistor increments is the Neural Engine, which accelerates neural network processing.
The A13 Bionic was 20% faster than the A12 Bionic's Neural Engine, but it had the same number of cores, eight. In A14 Bionic, however, the number has doubled to 16 cores and the computational power has increased by that amount. Whereas the A12 and A13 generations had 5 trillion and 6 trillion operations per second respectively, the A14 generation has dramatically increased its computing power to 11.8 trillion operations per second.
This massive increase in transistor resource allocation should be considered in conjunction with the features that will be included in the final product. However, there was little mention of this in the iPad Air announcement.
In other words, the next iPhone, which is expected to be called iPhone 12, may have some features with the greatly increased Neural Engine.
The Neural Engine is also used for Face ID facial recognition and improved camera quality. The iPad Air had Touch ID built into the suspend button, and the Neural Engine is used here as well. Today's mask-intensive society makes Face ID difficult to use, so it's possible that new elements of biometrics will be included.
Or perhaps camera quality and functionality, automatic photo and video classification and search, and automatic retouching processes will be built in beforehand. It's unnatural to allocate so many resources and yet have little to no mention of them.
Expect to see high performance versions of the A14 generation for the iPad Pro and Mac
By the way, the A13 generation did not develop an "X" version with more CPU and GPU cores and more memory bandwidth, respectively. This may be due to the close manufacturing process between the A13 and A12 generations. If the A13X Bionic were to be designed using the A13 generation design, the chip would have a significantly higher transistor count and would not be worth the cost.
However, with the Apple Silicon powered Macs coming this year, it wouldn't be surprising to see the iPad Pro get a full redesign in the spring of next year or so, as it needs to be differentiated from the iPad Air.
In addition, the manufacturing process has been scaled down to 5nm, so this generation will be designed with an X version. If that's the case, we'll probably see four high-performance cores in the CPU and eight in the GPU. This would result in a 40% CPU and 30% GPU speedup over the A12Z, more than double the speed of the Neural Engine, and 10 times the speed of the machine learning process.
These estimates are based on that ultra-thin chassis configuration used in the iPad Pro, so if you're going to use it in a MacBook Pro or similar device, you'll get a lot more performance.
The performance of Intel's 11th generation Core processor, the Tiger Lake, varies greatly depending on the thermal design of the on-board system. It's not surprising, since the performance of the cooling system and the onboard semiconductors are inherently linked.
The MacBook Pro should have more than twice the amount of heat generated by the SoC it can tolerate than the iPad Pro, so the performance of the Apple Silicon-powered Mac should be quite high. It will be interesting to see how it goes head-to-head with Tiger Lake, but I suspect that how the Neural Engine, ML accelerator, and ISP will be utilized on the Mac will also be reflected in the product's features and performance when it is announced.
This article is based on an article from the Japanese edition of Engadget and was created using the translation tool Deepl. The Japanese edition of Engadget does not guarantee the accuracy or reliability of this article.