Source: Content by Semiconductor Industry Watch (ID: icbank) original, author: Gong Jiajia, thank you.
It has been said that three apples changed the world, the first seducing Eve, the second awakening Newton, and Steve Jobs holding the third apple, changing people's communications, entertainment, and lifestyles. Whether it is the apple II, the first all-in-one machine with cross-era significance, or the first generation of iPhones that opened a new era of smartphones, Apple always seems to dare to be the "first person to eat crabs", and it is no exception in "core making".
When most mobile phone manufacturers have not yet "awakened" the consciousness of self-developed chips, Apple has begun to develop its own chips as early as 2007, through the acquisition of P.A semiconductors and Intrinsity, master the chip design technology. Jobs once believed that a company that really wants to do the best software must make its own hardware and do a good job of "software and hardware integration". The chip is obviously the most core hardware component, from the first Apple self-developed chip A4 landed in 2010, to the M1 Ultra released today, Apple has been running wildly for more than ten years on the road of "making cores", counting carefully, and there are more than thirty processors launched. In this article, the author will take stock of the chips (processors) that Apple has made over the years.
Because today Apple released the M1 Ultra, we start with this product and the series.
The M series that was born in the air
From the perspective of time, the M series should be the latest chip series launched by Apple, which is not difficult to understand, after all, compared with mobile phones, watches, and wireless chips, the design requirements of computer chips are obviously the most stringent. At present, Apple's previous series of M1, M1 Pro and M1 Max three products, plus M1 Ultra, Apple's M series has four chips.
According to Apple, the latest M1 Ultra is the next big leap forward for Apple chips and Macs. In the new chip, they used the UltraFusion package, Apple's innovative package architecture, to interconnect the dies of the two M1 Max chips to create a system-on-chip (SoC) with unprecedented levels of performance and functionality.
According to reports, the new SoC, also manufactured by the 5nm process, has 114 billion transistors, which is the largest ever in a personal computer chip. Configurable with up to 128GB of high-bandwidth, low-latency unified memory and accessible through a 20-core CPU, 64-core GPU, and 32-core neural engine, the M1 Ultra delivers amazing performance for developers compiling code, artists working in huge 3D environments that previously couldn't render, and video professionals who can transcode video to ProRes up to 5.6 times faster than those using an 28-core Mac Pro with Afterburner.
Specifically, in terms of CPUs, Apple's new chips now offer a total of 20 CPU cores. This includes 16 performance-focused Firestorm cores and 4 efficiency-focused Icestorm cores. Given that the M1 Ultra is only targeted at desktops, unlike the M1 Max, the performance gains are also noticeable.
Anandtech says that in practice, I would be surprised if the clock frequency of the M1 Ultra CPU core is much higher than that of the M1 Max. And they are mixed for Apple's CPU performance. For multithreaded workloads, they say, 16 Firestorm cores will provide enough throughput to top some performance charts. But for single-threaded/light-threaded workloads, Firestorm has been surpassed by newer architectures such as Intel's Golden Cove CPU architecture. So Anandtech stressed that we don't expect Apple to regain its lead in single-threaded performance here, we should focus more on machine translation, especially energy efficiency.
Doubling the number of M1 Max chips on a chip means that Apple is able to double the number of memory channels on the chip, doubling their overall memory bandwidth. According to Apple, the M1 Max has 16 LPDDR5-6400 channels with a total memory bandwidth of 408GB/s, while the M1 Ultra doubles it to 32 LPDDR5 channels and 800GB/s memory bandwidth. As with the M1 Max, this is achieved by soldering the LPDDR5 chip directly onto the chip package, with a total of 8 chips on the M1 Ultra.
Doubling the memory chip also allows Apple to double the total amount of memory available in its hardware. The M1 Max is up to 64GB, while the M1 Ultra is up to 128GB. But that's still less memory than on a truly high-end workstation like the Mac Pro.
Anandtech also points out that as seen at the launch of M1 Max, Apple has already provided its SoC with more bandwidth than the CPU cores consume alone, so double the bandwidth is unlikely to have much of an impact, rather than ensuring that the CPU cores are as good as they are on the M1 Max. Instead, all this extra memory bandwidth is designed to keep up with the growing number of GPU cores.
And GPUs are also what Anandtech thinks is the most interesting part of Apple's new chips. With 32 GPU cores, M1 Max has set a record for a monolithic integrated GPU. Now, Apple has doubled the number of GPU cores on a single chip to 64 GPU cores.
The Anandtech authors say that unlike multi-die/multi-chip CPU configurations, which have been common in workstations for decades, multi-die GPU configurations are a very different beast. The amount of internal bandwidth consumed by GPUs (well over 1TB/s for high-end components) has consistently made connecting them technically prohibitive. As a result, in traditional multi-GPU systems, such as mac Pro, where each GPU is presented as a separate device for the system, software vendors need to find innovative ways to use them together. In practice, this means having multiple GPUs handle different tasks, as insufficient bandwidth means they can't efficiently work together for a single graphics task.
But if you can somehow connect multiple GPUs to a huge die to die bandwidth — enough to replicate their internal bandwidth — then you might be able to use them together in a single task. This makes combining multiple GPUs transparently the holy grail of multi-GPU designs. Multiple companies have struggled to solve this problem for more than a decade, and Apple seems to be breaking new ground, becoming the first to achieve this goal.
As Apple says, the M1 Ultra is based on the extremely powerful and energy-efficient M1 Max. To build the M1 Ultra, the chips of the two M1 Max are connected using UltraFusion, Apple's custom package architecture. The most common way to extend performance is to connect two chips through a motherboard, which often presents significant trade-offs, including increasing latency, reducing bandwidth, and increasing power consumption. However, Apple's innovative UltraFusion uses a silicon interlayer to connect chips with more than 10,000 signals, providing 2.5TB/s of low-latency, interprocessor-to-processor bandwidth – more than 4 times the bandwidth of leading multi-chip interconnect technologies. This allows the M1 Ultra to run as a chip and be recognized by software, so developers can take advantage of its performance without rewriting code.
Regarding this package, according to Anandtech, it is a 2.5D package. Specifically, Apple designed an interface with very high speed on the edge of the M1 Max, and then, with the help of a silicon intermediation layer, allowed the connection of two M1 Max dies, about this technology, Apple did not disclose any details, but only emphasized the amazing speed of the above disclosure.
Anandtech also noted that while the details are very different from implementation to implementation, the foundation of the technology is the same. In all cases, some sort of silicon intermediation layer is placed under two chips, and then the signal between the two chips is routed through the intermediary layer. Silicon's ultra-fine manufacturing capabilities mean that a large number of traces can be routed between two chips — in Apple's case, more than 10,000 — which allows for ultra-wide, ultra-high-bandwidth connections between the two chips.
As mentioned above, with UltraFusion, Apple is able to provide an incredible 2.5TB/s bandwidth between two M1 Max chips. Even if we assume that this is a summary number — adding both directions together at the same time — it still means that they have 1.25TB/s bandwidth in each direction. This means that all of this is close to the internal bandwidth used by some chips and exceeds Apple's total 800GB/s total DRAM bandwidth.
As you can see, Apple has become the first vendor to bind two GPUs together and have such a large amount of bandwidth. This allows them to try to present two GPUs as a single device to the operating system and applications, as it allows them to move data quickly between GPUs as needed.
Thanks to this packaging technology, Apple said that the GPU performance of the company's M1 Ultra surpasses NVIDIA's GeForce RTX 3090, which is currently the fastest graphics card on the market. While doing this, the chip consumes just over 100 watts, which is a full 200 watts less than the RTX 3090.
The M1 Ultra also has a 32-core neural engine that can run up to 22 trillion operations per second to accelerate the most challenging machine learning tasks. In addition, the M1 Ultra's media engine capabilities are twice as powerful as the M1 Max, delivering unprecedented ProRes video encoding and decoding throughput.
The M1 Ultra also integrates custom Apple technologies such as a display engine capable of driving multiple external displays, an integrated Thunderbolt4 controller, and best-in-class security, including Apple's latest Secure Enclave, hardware-validated Secure Boot, and runtime-proof development technologies.
There is no doubt that this is an implementation that subverts everyone's inherent thinking. Then, let's look at the other products in the series.
M1
Launched in November 2020, the M1 processor is Apple's first self-developed chip for Mac computers, and in addition to the iMac and iPad Pro updated in 2021, it is also used in Apple's improved MacBook Air, 13-inch MacBook Pro, and entry-level Mac Mini.
M1 is based on TSMC's 5nm process, with up to 16 billion transistors, in the CPU part, M1 chip integrates 8 cores, including 4 high-performance large cores and 4 high-performance small cores, mixed operation to assist in multi-threaded tasks. According to Anandtech analysis, the big core of Apple's M1 chip uses the Firestorm microarchitecture, and the secondary cache of the CPU core is upgraded to 16M. Apple said that compared with the processor mounted on the Mac product in 2012, the CPU energy efficiency ratio of the M1 chip has been increased by 3 times.
In terms of GPUs, the M1 is equipped with an 8-core GPU, and the graphics performance is more than 5 times higher than that of the previous generation. In terms of NPU, M1 integrates a 16-core neural network processor (NPU) with a hash rate of 11TOPS. In addition, the M1 chip supports up to 16GB of unified memory architecture with high bandwidth and low latency.
M1PRO
As an upgraded version of the M1, the M1Pro is equipped with the 2021 MacBook Pro, which still uses the 5nm process, has 33.7 billion transistors, 10 core CPUs (8 high-performance cores, 2 energy-efficient cores), and is up to 70% faster, twice that of the M1. Among them, the large core adopts an ultra-wide pipeline architecture, each core is 192KB L1 instruction cache, 128KB L1 data cache, and the shared L2 cache doubles to 24MB with the number of cores. The small cores are a wide pipeline architecture, and each core continues to have 128 KB L1 instruction cache, 64 KB L1 data cache, and L2 cache 4MB, doubling each core on average.
GPU part, M1 Pro has 16 cores, 2048 execution units, supports up to 49512 concurrent threads, floating-point performance of 5.2TFlops, texture fill rate of 164 billion per second, pixel fill rate of 82 billion per second.
In terms of NPU, M1 Pro is also a 16-core, supporting hardware acceleration H.264, HEVCProRes/ProRes RAW video codec, and supporting multiple 4K and 8KProRes video streams.
In addition, the M1 Pro also uses a unified memory architecture with a total capacity of 32GB LPDDR5, 256-bit width, 200GB/s high bandwidth and low latency, and adopts a custom package designed by Apple.
M1MAX
M1Max is the most high-end version of the M1 series, equipped with the 2021 MacBook Pro, the process is still using the 5nm process, although it has the same 10-core CPU configuration (8 high-performance cores, 2 energy-efficient cores) with M1 Pro, compared to Pro, Max's graphics processing performance and memory bandwidth are all doubled.
According to reports, M1 Max memory bandwidth up to 400GB/s, RAM up to 64GB memory, GPU with 32 cores and 4096 execution units, concurrent threads can reach up to 98304, floating-point calculation force of 10.4TFlops, the fill rate on the texture is 3270 per second, and the GPU performance of the original M1 is increased by four times. In addition, the M1 Max has 57 billion transistors in an area of 432mm, the largest chip Apple has ever made.
The A series that won the first battle
The A series is Apple's small test knife on the road to making cores, from the A4 in 2010 to the A15 in 2021, the A series has launched more than 19 chips. At the same time, with the maturity of Apple's design process, the performance of A series chips is also multiplying like a rocket.
A4
A4 processor as Apple's self-developed chip "debut", according to the iPad official demonstration of the picture, the A4 processor manufacturing date is September 2009, officially released in January 2010, put into production in March, the first generation of iPad is its first product, since then began to appear on the iPhone 4, iPod Touch and Apple TV and other products. However, everyone seems to be less recognized by the self-research degree of this chip, because it has a similar core layout with Samsung S5PC110.
Specifically, the A4 processor uses a custom Samsung 45nm process 800MHz ARM Cortex-A8 single-core processor, at the same frequency, the CPU core can process about 10% more instructions than the ARM standard Cortex-A8 core, and the L2 cache is increased to 640KB. The GPU is the PowerVR SGX 535, which greatly improves the performance of the GPU by using the GPU core and RAM directly connected.
A5
As the first dual-core processor on the iOS platform, the A5 processor debuted at Apple's new product launch in March 2011, and was first installed on the iPad 2, with 2 times more computing performance and 9 times more graphics display performance than A4.
The A5 processor process is still Samsung's 45nm process, with a core area of up to 122mm2, a Multi-Core Cortex-A9 architecture processor, and a 9-layer copper interconnect layer + 1 polycrystalline silicon layer structure and stack package, supporting low-power DDR2 memory (LPDDR2) technology. The GPU core architecture is Imagination Technologies' PowerVR SGX543MP2.
A5X
Released in March 2012, the A5X is a new two-U fusion processor that runs on the iPad3 and has twice the graphics processing power of the A5 thanks to the integration of four graphics processors.
The A5X process is still Samsung's 45nm process, with a core area of 162.94mm2, equipped with a dual-core Cortex-A9, the GPU is SGX543MP4 250MHz, while the memory bit width is 128bit, using a quad-channel LNDDR2 32bit 800MHz, the bandwidth reaches 12.8GB/s, and has 2 times RAM to support its powerful 2048*1056 resolution screen.
A6 processor
Launched in September 2012 with the launch of the iPhone 5, the A6 uses Apple's first-generation self-developed CPU core, which is twice the performance of the previous generation.
The A6 is based on the ARMv7 instruction set, using Samsung's 32nmHKMG process, dual-core 1.3ghz frequency, with higher performance and lower power consumption. On the GPU side, the A6 integrates a three-core PowerVRSGX 543MP3 graphics processing unit with a memory bit width of 64-bit and four USSE2 pipes, each with four vector ALU units and 32Flops floating-point performance.
A6X
Released in October 2012, the A6X runs on the iPad4 (2012), integrates two CPU cores and four GPU cores, and delivers twice as much CPU and graphics performance as the A5X processor.
The A6X is manufactured using the Samsung 32nm HKMG process with a core area of up to 123mm2, 30% higher than the A6. The CPU part is basically unchanged, almost exactly the same as the A6 layout, two Swift cores designed by Apple itself, but the main frequency has been increased from 1.3GHz to 1.4GHz. The A6X's GPU cores are increased from 3 to 4, and each GPU core is divided into 9 sub-cores, of which 4 are a group, two sets of the same set of sub-cores, plus a central core. This processing will increase the maximum clock frequency, giving the iPad 4 better graphics power.
A7 processor
The A7 is the world's first 64-bit mobile SoC, released in 2013, first applied to the iPhone 5s, in addition to the iPad Mini 2 (2013), iPad Mini 3 (2014). According to Apple, the A7 is 2 times the performance of the A6, 40 times that of the original generation, and 56 times the graphics capability of the original.
The A7 processor is a dual-core processor based on a 28nm process (ARM v8 architecture, clocked at 1.3GHz), with a built-in GPU of quad cores (Power VR G6430), a core area of 102mm2, and a transistor count of about 1 billion, providing a secure Enclave for protecting Touch ID biodata. In addition, there is at least 3MB of SRAM (static random access memory) in the safe zone in the A7, which does not require a circuit refresh to save the stored data).
A8 processor
The A8 processor was launched in September 2014 and runs on iPhone 6, iPhone 6 Plus, iPad Touch (6th generation), iPad Mini 4, Apple TV (4th generation), homePod. According to the official announcement, the A8 processor can get up to 25% CPU performance improvement and 50% graphics performance improvement.
The A8 processor uses a 20nm process and contains 2 billion transistors, and although the A8 has twice the number of transistors as the A7, the physical size is reduced by 13% to 89 mm2. The CPU uses the second generation Cyclone core, the main frequency is 1.4GHz, the GPU is 4 PowerVR G6450 cores, based on six USC (Unified Shader Cluster) and PowerVR's own Rogue architecture.
A8X
Released in October 2014, the A8X was Apple's first triple-core mobile processor, and Apple claimed that the A8X was 40% more powerful than the Apple A7.
The A8X processor chip area is about 125mm2, the total number of transistors is about 3 billion, compared to the A8, the number of transistors increased by about 50%, but the area increased by only 40%. In addition, A8X uses a 20nm process, using ARMv8.0-A for a triple-core CPU, and more built-in eight GPU cores, the graphics processor is PowerVR GXA6850, and it is estimated that the area of the GPU part of GXA6850 is about 38mm2, accounting for about 30% of the entire processor.
A9 processor
Released in September 2015, the A9 processor is Apple's third-generation 64-bit dual-core mobile processor for the iPhone 6s, iPhone 6s Plus, iPhone SE, and iPad 5th. Apple claims that the overall performance of the A9 is as much as 70% higher than that of the previous generation, and the graphics performance is as high as 90%.
It should be noted that the A9 processor is divided into two versions by Samsung or TSMC, of which the A9 processor model for iPhone 6s is APL0898, and the package is Samsung 2GB LPDDR4 RAM, which is manufactured using the Samsung 14nm FinFET process; and the A9 processor model for iPhone6s Plus is APL1022, which is packaged with Hynix 2GB LPDDR4 RAM and TSMC 16nm FinFET process manufacturing.
Although the core area of the two chips is about 9%, the internal structure is almost the same, with the same CPU, GPU, memory controller, coprocessor, etc., but there will be some slight differences in the specific layout and details.
Taking the Samsung version as an example, this A9 chip contains a dual-core CPU, a six-core GPU and two SRAM, as well as a coprocessor M9. Among them, the CPU part is based on the ARMv8-A architecture, and the main frequency is increased to 1.8GHz. In addition, the A9's L1 cache is very large, with each core having 64KB of L1 data cache and 64KB of L1 instruction cache, with a L2 cache of 3MB and a L3 cache of up to 8MB.
In terms of GPUs, the A9 uses PowerVR's latest seventh-generation PowerVR GT7600 GPU. The GT7600 has 6 USCs, each with 32 ALU, so there are a total of 192 ALU cores.
A9X
Launched in September 2015, the A9X processor is the one used in the iPad Pro, and the A9X has twice the bandwidth and storage performance compared to the previous A8X processor. The chip is 1.8 times faster than the A8X, and the GPU performance is 360 times faster than the first-generation iPad.
A9X core area of 147mm2, using TSMC 16nm FinFET process, dual core, core architecture for ARMv8.0-A, GPU for PowerVR Series 7XT series, a total of 6 GPU units in the core, each unit has two GPU cores, that is, a total of 12 GPU cores, a total of 384 stream processors.
A10 Fusion
The A10Fusion, Apple's first quad-core processor, was officially released in September 2016, which is a 40% improvement in processor performance compared to the A9, 120 times that of the first-generation iPhone, and 50% higher than the A9 processor in graphics processing. It is equipped with the iPhone 7, iPhone 7 Plus, iPad (sixth generation), iPad (seventh generation) and iPod touch (seventh generation).
A10Fusion has 3.3 billion transistors, based on the ARM architecture, using big. The FOUR-core SoC configured by LITTLE consists of two high-performance cores and two energy-efficient cores. Among them, the high-performance CPU core is codenamed "Hurricane" (Hurricane) and the low-power CPU core codenamed "Zephyr", both of which are ARMv8-compatible microarchitectures designed by Apple itself. The GPU is the PowerVR GT7600, with a core area of 125mm2 and A TSMC 16nm process.
A10X Fusion
The A10XFusion was released in June 2017 and runs on the iPad Pro 2 (10.5-inch 2017), iPad Pro 2 (12.9-inch 2017), and Apple TV 4K. According to the official website, the A10X is 30% and 40% higher than the A9X in CPU and GPU performance.
The A10X uses TSMC's 10nm process and has a core area of 96.4mm2, which is about one-third smaller than the A9X area. The A10X uses a new "3+3+12" architecture, of which 3 are high-performance processing cores, 3 are energy-efficient processing cores, and the remaining 12 are GPU processors. The CPU part is designed for 6 cores, 3 large and 3 small, with a frequency of 2.36GHz and 128-bit LPDDR4 memory.
A11 Bionic
Launched in September 2017 on the iPhone 8, iPhone 8 Plus and iPhone X, the A11Bionic is the world's first processor with a Neural Network Engine (NPU). Apple says the A11 is 70% more energy efficient than the A10 and has a 25% improvement in performance.
The A11 processor integrates 4.3 billion transistors with a core area of 87.66mm2 and uses TSMC's 10 nm process. In terms of CPUs, the A11 is equipped with a 6-core CPU with a 64-bit ARMv8-A architecture, including 2 performance cores called "Monsoon" and 4 energy efficiency cores called "Mistral", which are 25% faster than those in the previous generation A10, and 70% faster energy efficiency cores. In terms of GPUs, the A11 is powered by Apple's first independently designed triple-core GPU, which Apple says is 30% faster than the Imagination GPU used on the iPhone 7.
The highlight of the A11 is that it is equipped with a hardware dedicated to machine learning, the "neural engine". The A11's neural engine features a dual-core design with up to 600 billion operations per second, equivalent to 0.6TFlops, specifically for FACE ID, Animoji, and ASICs for AR applications.
A12 Bionic
Launched in 2018, the A12Bionic is the industry's first 7nm process chip on the iPhone XS, iPhone XS Max, iPhone XR, iPad mini (fifth generation), iPad (eighth generation), iPad Air (third generation), and Apple TV 4K (second generation).
It is understood that the 6-core cpu, 4-core graphics processor and 8-core neural network engine contained in A12 Bionic are all developed by Apple itself. The A12 Bionic uses two Vortex 2.5GHz large cores and four 1.59GHz Tempest small cores to form a six-core CPU, and by adopting a six-core fusion architecture, the speed of two performance cores can be increased by up to 15%, and the energy savings of four energy-efficient cores can be up to 50%. The GPU is a self-developed quad-core G11P with a frequency of more than 1.1GHz and a graphics performance increase of up to 50%.
In addition, the A12 Bionic Neural Network Engine is equipped with eight cores and can perform five trillion operations per second, and the Core ML running on the A12 Bionic Neural Network Engine can be up to nine times faster than the A11 Bionic, but it consumes one-tenth the energy consumption of the A11 Bionic.
A12X Bionic
The A12XBionic was released in October 2018 and is available on the iPad Pro 3 (11-inch 2018) and iPad Pro 3 (12.9-inch 2018). In terms of CPU performance, Apple said that the A12X's single-core performance is 35% higher than that of the A10X, and multi-core performance is 90% higher.
The A12XBionic also uses the 7nm process, but the number of transistors has increased from 6.9 billion to 10 billion A12. In terms of CPU, it is equipped with Apple's first 8-core CPU architecture, which consists of 4 Vortex high-performance cores and 4 Tempest energy-efficient cores. In terms of GPUs, the A12X processor is a 7-core GPU architecture. In terms of NPU, the A12X also supports NPU units on the A12, and the performance is also 5 trillion times, which should also be an 8-core NPU unit.
A12Z Bionic
The A12ZBionic was launched in March 2020 on the iPad Pro 4 (11-inch 2020) and iPad Pro 4 (12.9-inch 2020).
According to media reports, the A12Z and the 2018 iPad Pro use the A12X processor has not been much improved, and the processor architecture is almost no different. Apple said in a press release released on the new iPad Pro that one of the changes to the A12Z Bionic processor is the addition of an eighth GPU core.
In fact, the A12X Bionic also has eight cores, but one of them is disabled by Apple, and the A12Z only "denucleates" the blocked core.
A13 Bionic
The A13Bionic was released in September 2019 and is available on the iPhone 11, iPhone 11 Pro, iPhone 11 Pro Max, iPhoneSE (2020), and iPad (2021).
The A13Bionic uses the second generation 7nm process, containing 8.5 billion transistors, 6 CPU cores consist of 2 high-performance cores running at 2.66 GHz (called Lightning) and 4 efficiency cores (called Thunder), 2 high-performance cores are 20% faster, 30% slower, 4 performance cores are also 20% faster, and the power consumption is reduced by 40%.
In terms of GPUs, the A13 GPU is a quad-core design, which is 20% faster and 40% lower in power consumption. At the same time, the A13 also has an 8-core neural computing engine, which improves performance by 20% and reduces power consumption by 15%.
A14 Bionic
Launched in September 2020, A14Bionic is mounted on the iPad Air (fourth generation), iPhone12, using TSMC 5nm process, integrated 11.8 billion transistors, the CPU is a 2 large core + 4 small core design, two high-performance cores for handling complex tasks, the remaining four CPU cores for general performance cores, designed for general work, but performance increased by 40%. The GPU continues to maintain 4 cores, but with a new architecture upgrade, performance is improved by 30%.
As for the AI core, the AI core has been increased from the previous 8-core to the 16-core architecture, and the AI computing power has increased to 11.8 trillion times, claiming that the performance is twice that of the previous generation.
A15 Bionic
The A15Bionic is the latest chip in the A series to date, released in September 2021, on the iPhone 13 mini, iPhone 13, iPhone 13 Pro, iPhone 13 Pro Max, and iPad mini (sixth generation).
The A15Bionic uses TSMC's latest 5nm process, integrates 15 billion transistors, has 6 cores, and includes 2 high-performance large cores + 4 energy-efficient small cores. It is understood that the A15 Bionic also has two versions, of which the A15 Bionic chip on the iPhone 13 and iPhone 13 mini has a 6-core cpu (2 performance cores and 4 energy efficiency cores), a 4-core graphics processor and a 16-core neural network engine. The A15 Bionic chip on the iPhone 13 Pro, iPhone 13 Pro Max and iPad mini (sixth generation) has an additional 1-core graphics processor.
Step-by-step S series
The S series is a chip developed by Apple for smart wearable watches, and from the perspective of time, the S series was launched 5 years later than the A series. Due to the small size of the Apple Watch and strict restrictions on energy consumption and heat dissipation, Apple designed SiP on the basis of the old A series chip, and at present, the S series chip has been iterated to the S7 chip.
S1
Released in April 2015, the S1 runs on the Apple Watch (Gen 1) with a core area of 26mm*28mm and 30 separate components, including NXP's NFC chip, AMS's NFC signal amplifier, and Maxin's audio amplifier.
It is understood that the S1 chip uses Samsung's 28nm LP process, the maximum CPU frequency is 520 MHz, and the GPU use may be PowerVR Series 5 GPU.
S2
Released in September 2016, the S2 runs on the Apple Watch Series 2 in an S2 System-In-Package (SiP) with a dual-core CPU that is 50% faster than the product model at the time and a GPU that is twice as fast.
S3
The S3 was released in September 2017 and runs on the Apple Watch Series 3. Its dual-core processors are divided into cellular and non-cellular versions, and the cellular version of the package is the same as before, but the non-cellular version of the S3 processor is a conventional plastic BGA package. The S3 processor is close to the S2 processor in terms of chip area, which is greatly reduced compared to the S1 processor, reducing power consumption and improving performance.
S4
Released in September 2018, the S4 runs on the Apple Watch Series 4, which is dual-core but upgraded to a 64-bit architecture, which is twice as fast as the S3.
S5
The S5 was released in September 2019 and runs on Apple Watch Series 5 and Apple Watch Series SE. The S5 chip uses a 64-bit dual-core processor, which is 2 times faster than the S3 processor, including the W3 wireless chip.
S6
The S6 was released in September 2020 and runs on the Apple Watch Series 6. According to reports, the S6 chip uses a sixth-generation package module, containing a high-performance dual-core processor, which is based on the A13 bionic of the iPhone 11, optimized for the Apple Watch, and the speed of the S6 is 20% faster than the previous generation.
S7
Released in September 2021 and running on the Apple Watch Series 7, the S7 is a 64-bit dual-core processor that integrates GPU, 32GB flash, Bluetooth 5.0, 1 GB of RAM, 802.11 b/g/n 2.4 and 5 GHz WiFi, and satellite positioning (GPS, GLONASS, Galileo, QZSS).
It is understood that the S7 has two CPU cores based on the Apple A13, which are 20% more powerful than the S5 and are technically the same as the S6 (sama model t8301).
Parallel W and H series
The wireless chip series was launched a year later than the S series, and the current wireless chip series contains two series, W and H, of which the W series has released three products, and the H series has only released 1 model.
W1
The W1 chip, which debuted in 2016 as Apple's first wireless chip, is about 14.3mm2 in size and is mounted on AirPods, which maintains a Bluetooth 1 connection with computer devices and decodes the audio streams sent to it.
W2
The W2 chip was launched in 2017 and is equipped with the Apple Watch Series 3, which supports Bluetooth and has been integrated into the Apple S3 SiP. Apple said that compared with the previous chip, w2 has increased Wi-Fi speed by 85%, and Bluetooth and Wi-Fi power efficiency by 50%;
W3
The W3 chip was launched in 2018 with Bluetooth 5.0 connectivity and is powered by the Apple Watch Series, which has been integrated into the Apple S4, S5 and S6 SiP.
H1
Launched in March 2019 with a core area of 12mm, the H1 has the same performance as apple's iPhone 4 (powered by an A4 processor), allowing the headphones themselves to perform a large number of tasks, reducing latency and improving connectivity performance. In addition, the H1 chip also supports Bluetooth 5 Class 1 and activateSiri with voice.
Write at the end
As I said at the beginning, we've only covered the processors Apple has developed over the years in this article. But in fact, so far, the territory of Apple's self-developed chips has expanded to power management, screen drivers, Bluetooth headsets, basebands, fingerprint recognition, 3D somatosensory and other fields.
With the further strengthening and improvement of Apple's ecological chain in the future, I believe that Apple will go farther and farther on the road of "making cores", and will also bring us more self-developed chips.