This article is based on an article from the Japanese edition of Engadget and was created using the translation tool Deepl.
The Mac with Apple's proprietary processor. What will the processor be?
We've been paying attention to the Mac itself, but if we are looking for a way to find out what it means to be proprietary, the focus will inevitably be on the performance of the processor. And the actual product announced makes the Apple Silicon, or Apple M1 chip for the Mac, even more compelling.
Described as "compared to the latest laptops", Apple announced that the CPU and GPU will both perform twice as well when set to 10 watts of power consumption. It also said that the CPU and GPU can deliver "the same maximum performance as the latest laptops" at 25 percent and 33 percent of the power, respectively.
These latest laptops are laptops powered by the Intel Core i7-1165G7, which is the latest model to use the 11th generation Intel Core and Xe graphics architecture. If their claims are accurate, it will have a significant advantage over other laptops in terms of performance per power consumption.
The newly announced MacBook Air, MacBook Pro, and Mac mini all look exactly the same as their predecessor models. I suspect this is to give a sense of the overwhelming difference between them and the Intel version. That's how big the performance difference between the new chips and the old ones is. Even though they are all in the same chassis, the MacBook Air has more performance than the Intel version without the cooling fan. The MacBook Pro and Mac mini with a cooling fan do not close the gap in scalability. If they had changed the design, the difference between the two would not have been as pronounced.
We'll get into the details of each product in a later review of the actual machines, but for now, let's focus on the M1 and see what it brings to the Mac.
Why the M1 is not the A14X Bionic
We thought the first Apple Silicon Macs would come with the A14X Bionic - an enhanced version of the A14 Bionic with more CPU and GPU cores. But what appeared was an "M1" chip. Supposedly, the name A14X will be used for the new iPad Pro, but there's actually a reason why it's called M1.
The M1 is not an extension of the A14 Bionic, it is a reconfiguration of Apple's latest circuitry and re-packaged for the Mac. Of course, many of the latest circuits (components) developed by Apple are the same as those developed for the A14 Bionic. While the high performance and efficiency cores, 16 Neural Engine cores, GPU cores, ISP, etc. are identical in design, the M1 has four of the same high-performance cores and eight GPU cores, each doubling the number of cores in comparison to the A14 Bionic, and doubling the GPU cores results in double the memory bandwidth. These are similar to the relationship between A12 Bionic and A12Z Bionic.
However, Macs require some features that are not needed on iPhones or iPads. One of them is virtualization, and the M1 adds the ability to run a virtual computer (but not x86 processor virtualization. The instruction set is ARM only). The GPU architecture is the same, but it is said to support more texture formats. This will be important in supporting games for the PC, for example. Likewise, what the A14 Bionic does not have is the Thunderbolt 3/USB 4 interface.
In other words, the M1 takes the design developed for the iPhone/iPad and uses it as a component, but it also has all the elements needed for Macs in the same die. It's unclear whether the M1 will be used for future iPad Pro model changes or if the A14X Bionic will be built separately, but either way, the M1 is definitely a dedicated processor with elements unique to Macs.
Performance dynamically changes with thermal design; the M1 has only one specification
All three Macs that have been introduced this time around have a single M1, but with the exception of the entry-level MacBook Air model, which has a limited number of GPU cores. Every model has an M1, with no clock frequency listed. That's because the maximum operating clock frequency is the same for all of them (it's not publicized and cannot be known except by measuring the actual device). To be precise, the range of operating clock frequencies is the same.
Like the latest 11th generation Intel Core, the M1 dynamically changes its operating clock depending on the cooling performance of the system on which it is installed, the operating environment (e.g., temperature), and the operating conditions of the application. In other words, whether it's a fanless, thin MacBook Air or a mini-desktop Mac mini, the M1 itself is the same and can be considered nearly identical in terms of instantaneous power. But if you're running a multithreaded, core-full, GPU-intensive game or application, the cooling capacity of the Mac mini is more than enough to get the performance out of it.
In today's world of processors where performance is governed by power, this kind of configuration is quite conceivable. From a different perspective, however, this kind of usage is only possible because Apple is a company that develops and sells PC systems.
Intel and other CPU vendors have been selling finished chips at different prices for different ranks of chips, based on the selection and classification of their performance. However, for Apple, which only sells end products, there is little point in creating a price lineup based on such sorting. Instead, it would be more natural for performance to vary depending on the design.
High performance and limitations because of shared memory SoC
The reason for the M1's low power consumption and high performance is because of the integrated system chip (SoC) architecture that has been refined for iPhones. The company has been working in-house to develop the processing circuitry that goes into this SoC for the iPhone and has perfected a system architecture in which different processing circuits share the same memory and work together. The unified memory architecture (UMA) is extremely efficient for running CPUs, GPUs, and a dedicated processor (Neural Engine) for inference algorithms in machine learning models in parallel.
In particular, when using the GPU for general purpose, it is not very efficient to load data into video memory to reflect the results of GPU processing, process it, and draw it back into memory. It would be more efficient if the same data could be accessed by a CPU, a GPU, a Neural Engine, an ISP, and so on, each with its own specialties. It is clear which is more efficient: having multiple specialists processing information around the same table, or having them take turns exchanging information via email.
Higher efficiency means better power consumption. The good design itself and the detailed power management functions introduced in the A13 generation may have helped, but UMA must have also contributed to the reduction in power consumption. Another advantage of UMA may be the ability to increase performance by increasing the number of cores, simply by increasing the interconnection bandwidth within the SoC.
However, UMA has its own limitations. The memory bandwidth shared by each processing circuit must be large enough to achieve performance gains with UMA. The reason why the M1 encapsulates DRAMs in a single package in the form of a SIP (system-in-package) is presumably to maximize the bandwidth to memory.
Why Apple is going Apple Silicon, starting with its low-end models
In other words, it can only have as much main memory as it can fit in the same package, and this time, the economically reasonable line was probably 16GB maximum. Whether or not Apple M processors will be used in even higher-end products in the future, it seems that improving the main memory capacity will be essential.
Similarly, the integration of even the I/O interface into the chip is an advantage, but it can also be a weakness; the M1 only supports two Thunderbolt 3/USB 4 ports. This is expected because although it incorporates an interface within the M1, it only supports two channels. The concept of consolidating a large portion of a system into a single package is advantageous in terms of performance, power consumption, and security, but it also raises the bar for building a larger system.
For this reason, Apple is moving all MacBook Air models to the M1, but only the lower 13-inch models (Thunderbolt 3x2 port models) in the MacBook Pro series, and the new Mac mini will be sold alongside the Intel-powered model with four Thunderbolt 3 ports. All of them also differ in terms of maximum memory and SSD capacity that can be installed.
Considering the simple CPU/GPU performance, and the significance of the Neural Engine, it seems like a bold move to Apple Silicon, however, the reason for the introduction of Apple Silicon in part of the lineup, and more specifically in order of the lower-spec models, is probably because of system scalability constraints in all aspects of the system, not just CPU/GPU performance.
Apple has said that it will take two years to revamp its lineup, and it will undoubtedly be some time before the Mac Pro is replaced with an Apple silicon-powered machine.
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.