In today's mobile processor space, Apple is undoubtedly the number one player in the performance project. At the iPhone 13 launch, Kaiann Drance, Apple's vice president of product marketing, said bluntly that friends are still struggling to catch up with Apple's chips two years ago. Judging from the benchmark achievements and actual experience, this is an indisputable fact.
Why did Apple do this? Today we will talk about the key factors that make Apple chips stronger.
Large cache, only this one
Friends who often toss computers should have heard of a concept: the larger the processor cache (Cache), the better the performance. This statement is not accurate, but it is also half right, and it is more accurate to say that the larger the cache, the better the performance when the processor structure and process are similar.
In principle, the processor cache is a temporary data exchanger between the CPU and memory, which is generally integrated inside the processor to solve the contradiction between the CPU processing speed and the memory read and write speed. Modern computer processors generally have L1, L2, L3 level three cache, L1 cache is closest to the CPU, the fastest, the smallest capacity, L3 cache speed is slower (relatively speaking, in fact, not slow at all), the largest capacity. The speed here is relatively speaking, no matter what level of cache, the speed is faster than the memory and hard disk.
The relationship between memory, multi-level caching, and CPU
Apple's self-developed processors have always favored large cache designs. Starting with the A7 processor on the iPhone 5S in 2013, Apple modeled on the desktop processor and pioneered the use of L3 cache on smartphones, with 1 MB L2 cache and 4 MB L3 cache, and the A8 cache specifications are the same. After the A9 and A10 processors, Apple used 3 MB L2 cache and 4 MB L3 cache.
And what about Android phones? ARM didn't add L3 caching to its public architecture until the Cortex-A75 generation, up to 4MB, in the summer of 2017. The first Qualcomm processor platform with cortex-A75 core is the Snapdragon 845, which was not available until the winter of this year, with 2MB L3 cache (not full), and when the mobile phone equipped with the chip is listed, it is 2018 again.
On the other hand, Apple's 2017 A11 chip eliminated the L3 cache, but increased the L2 cache to 8MB, more than the Snapdragon 845 three levels of cache combined.
In 2019, Apple A13 was unveiled with the iPhone 11 series, maintaining the 8MB L2 cache specification, and for the first time introduced the System Level Cache as the last level cache of the processor, with a capacity of 16MB, which can act as a shared cache for all IPs on the chip, similar to the Infinity Cache of AMD graphics cards.
On the Android platform side, Qualcomm released the Snapdragon 865 after Apple, and also introduced a system-level cache design. However, Qualcomm is still not bold enough this time, and the Snapdragon 865 system-level cache is only 3MB, which is far from the 16MB of the A13. In addition, the L1, L2, and L3 caches of the Snapdragon 865 add up to only 6.8MB, which is not as large as the L2 cache of the A13.
By the iPhone 13 generation, the cache capacity of the full-blood A15 has approached the desktop processor level, the performance core L2 cache has increased to 12MB, the same M1, and the Snapdragon 888's L2+L3 cache is only 6.5MB in total; the system-level cache has increased to 32MB, while the Snapdragon 888 is only 3MB, and Huawei's Kirin 9000 heap to 8MB is still much behind the A15.
Without him, willing to spend money
The big cache is so direct to the performance improvement, why do manufacturers outside of Apple not like to use it?
This is because Apple's chips are "bigger" than other manufacturers' ones. One is large in the physical sense, that is, the chip area; the other is large in the sense of price, that is, the cost is higher. The two are complementary to each other, just as the so-called chip space is scarce, the larger the chip occupies space, the higher the cost.
Apple M1 Ultra vs AMD desktop processor
According to a 2016 report by the Linley Group, the Apple A10 chip has about twice the performance core area of the competitor's high-end mobile CPUs, as well as its energy-efficient core, which is nearly twice that of the competitive processor's energy-efficient core Cortex-A53 in the same period.
Turning to the cache configuration we talked about earlier, after Apple added system-level caching on the A13, the SRAM area on the chip directly doubled. Apple's price tag in the processor cache is getting bigger and bigger, which not only directly increases the cost of materials, coupled with a large number of changes to the public processor architecture, the requirements for process and design capabilities are also getting higher and higher, which further increases the cost of chips.
Apple can do this because it sells finished phones, not chips. The chips made by Apple are only used by their own products, and they are not sold to other manufacturers, so they do not have to consider too many cost factors. The two major SoC suppliers of Android mobile phones are Qualcomm and Samsung, in order to sell chips, Qualcomm and Samsung inevitably have to bargain with mobile phone manufacturers, and they have to consider the cost of chips.
This can be said to be Apple's unique advantage. According to Counterpoint, Apple's iPhone controls about 70% of the world's profits with less than 20% of smartphone shipments, and more than 80% in many cases. With such high profits, Apple can invest heavily in chip research and development and manufacturing.
The revenue and profit of Apple's mobile phones are one dimension higher than those of friends
Apple is really too profitable to make mobile phones, what is the point of spending more money to make chips?
An apple that is one step ahead
From the A4 in 2010 to the A15 in 2021 to the M1 Ultra this year, Apple's self-developed chip process is only a short 12 years, which is not a long time for the chip manufacturing industry, and the seniority of Apple's competitors is much longer than this.
Founded in 2004, Huawei's HiSilicon released the first generation of self-developed chip K3V1 in 2008; Qualcomm announced its entry into the field of mobile phone processors in 2005, and its feature machine processor platform was first launched in 2006; Samsung built a smartphone processor S3C44B0X for Danger Hiptop in 2002, and has been engaged in the research and development of ARM media player chips in the 1990s; Intel, AMD, Nvidia, IBM, focusing on the PC field. There is also Motorola, which has withdrawn from the processor market, all of which are representatives of decades of experience.
Danger Hiptop with Samsung processor (2002)
Unlike other opponents, Apple has taken several big strides in the past 12 years.
Before the launch of the first self-developed processor chip, Apple made a series of blockbuster acquisitions, the most famous two being the acquisition of PA Semi in 2008 and the acquisition of Intrinsity in 2010. The core-making history of these two companies, which can be traced back to the 1980s and 1990s, is an old seniority in the chip industry.
After embarking on the road of self-developed chips, Apple has not stopped the pace of exploration. As early as 2013, on the A7 processor, Apple used a 64-bit processor architecture. Qualcomm dismissed this, with one executive declaring that the 64-bit A7 processor was just a gimmick. A few months later, Qualcomm consciously started a 64-bit processor, one step slower.
From the beginning of its entry into 64-bit processors, Apple has seized the lead and has maintained it to this day.
It is worth mentioning that ARM launched a program in 2016 to give chip partners earlier exposure to their latest CPU designs, from design and tuning. Judging from the pace of Apple's self-developed processor, perhaps Apple has been doing this for a long time.
Performance is important, but not all
When it comes to the performance of smart phones, we will inevitably touch on the "performance surplus theory", that is, today's smart phone performance is enough, and no matter how strong the performance is, it is also excessive. In fact, whether it is a smartphone or a computer, the hardware performance will not be enough, at least for the foreseeable time.
There's a famous Andy-Bill law in computing, "As much As Andy gives, Bill takes as much." "The gist is that the performance gains brought about by the new hardware will be consumed by the new software. On the other hand, developers' software updates are built with the performance of new hardware in mind, and will consume more hardware performance. When buying a mobile phone or computer, the stronger the hardware performance of the device, of course, as long as possible.
Former Intel CEO Andy Grove and former Microsoft CEO Bill Gates
From another point of view, the emergence of the "excess performance theory" also tells us that after the hardware performance can meet the daily use, people began to pay attention to the multi-dimensional experience outside the traditional CPU and GPU performance, such as AI performance, graphics processing performance, network rate, etc., which is closely related to the experience of map recognition ability and camera performance.
Perhaps, people's indifference to performance is the embodiment of the high level of hardware performance.
『Hot Recommendation』