In November 2020, Apple announced its first generation of M-series computer chips in the Apple Silicon series of in-house-designed chips for Mac computers, replacing Intel central processing units (CPUs). The M1 chip, now built into the latest MacBook Pro, iMac, and iPad Pro models, has proven itself competitive within the laptop market, and Apple appears confident in the potential for Apple M-series chips to outclass even high-end chips used in desktop PCs. The Apple Silicon transition plan revealed the company’s eagerness to invest in disruptive hardware technologies for its computers, as well as its lower performance expectations of Intel chips for use in such propelling products.
Are engineers at Apple correct in their predictions that M-series chips are capable of passing higher benchmarks and beating out competitors such as Intel and AMD, currently and into the future? In this article, we will look at the Apple M1 and other M-series chip’s hardware features, manufacturing processes, performance benchmarks, and comparisons to its competitors to better understand the Apple Silicon project and how it reflects current trends in the semiconductor industry and the wider PC market.
Components of the Apple M-series chips: SoC, not just CPU
One key design feature of Apple M-series chips sets them apart from chips previously used in Mac computers. While a “chip” in this context is often used synonymously with CPU or processor, the Apple M1 chip integrates several components including the CPU, GPU, and parts that control I/O, security, and other aspects, as detailed in Apple’s 2020 press release. Combining these technologies into one central component with a unified memory architecture, or shared memory pool, allows for faster and more energy efficient communication across all parts of the system. Traditionally, desktop PCs and laptops have kept these components separate, which requires processors in a system to redundantly copy data between several memory pools and transfer them over a larger area.
Image: Unified memory architecture.
Source: How “Unified Memory” Speeds Up Apple’s M1 ARM Macs.
The M1’s 8-core CPU contains four high-performance (P) cores and four high-efficiency (E) cores. The M1 Pro and M1 Max, launched in October 2021, contain eight P and two E cores for a 10-core CPU. P cores are utilized for intensive computing processes such as video editing and software development, while E cores are used for lighter processing tasks such as web browsing. This asymmetric CPU design allows applications to balance performance and efficiency and is the first processing unit for general purpose computer systems, inarguably an impressive milestone achieved by Apple.
Although core allocation is determined by the operating system through a Quality of Service (QoS) control system, which works well for some processes, for others, it appears far from fully optimized when subjected to testing. Alongside core counts, the actual results of this core allocation process is something to consider when judging asymmetric CPUs.
Architecture of M-series CPUs: ARM over x86
Apple’s M-series chips are the first ARM-based chips used in Mac computers, as opposed to Intel’s x86-based chips. The key difference between these two architectures lies in their instruction sets. ARM devices incorporate a simpler instruction set that allows them to be smaller, less energy intensive, and less heat generating — perfect for inside Apple’s staple mobile devices, and more recently, for laptop computers.
While cheaper to manufacture than standard PC processors such as Intel’s, ARM processors are generally less powerful computationally and have long been considered viable only in the mobile device market. However, ARM technology has recently reached a high enough performance level to be competitive with x86 processors in the mainstream laptop market, with Apple Silicon being the first higher end chip series to signal this potential market shift.
ARM is also a semi-open architecture, while Intel’s x86 is closed source. The developers of ARM license their CPU design for use by companies such as Apple, and those companies are then able to design their ARM chips to be highly compatible with the rest of the system and its software. M-series chips are custom-built to optimize CPU performance through tighter hardware-software integration, one of the main reasons why Apple Silicon is able to mitigate performance drawbacks brought by ARM architecture as compared to x86-based chips.
Manufacturing Apple Silicon: TSMC’s 5-nanometer process
M-series chips are made by TSMC, a Taiwan-based contract manufacturer of computer chips, using leading-edge 5nm semiconductor lithography process technology, which is often touted as being far ahead of Intel’s current 10nm processes. Historically, these measurements have referred to the size of each transistor on a chip, and the smaller the number, the more efficient the chip.
Nowadays, the numerical reference of lithographic processes are nothing more than marketing labels and do not accurately map onto the physical features of a chip. Other parameters, such as transistor density (million transistors per square millimeter), must be accounted for when comparing chips, especially chips made by different manufacturers.
Intel’s planned 7nm process is said to be comparable to TSMC’s 5nm process, while some sources report estimates of Intel’s 7nm transistor density being much higher than that of TSMC’s 5nm (though Intel recently changed it’s process node naming scheme to better match that of TSMC). It’s an industry complicated by evolving technology and terminology, though we do know that TSMC is ahead of Intel in regard to the processes currently used in production. Intel 4, formerly known as Intel’s 7nm process, is set to arrive in 2023, after being delayed almost two years following manufacturing issues. TSMC has been making 5nm chips since 2020, and expects to begin mass production of 3nm chips by 2023.
Images: 5nm and 7nm node densities. Source: WikiChip.
Benchmarks: How does Apple Silicon compare?
There are a number of benchmarking strategies designed to test performance differences between computer chips. In-depth discussion of these strategies is beyond the scope of this article. However, from a high level overview, Apple’s M-series chips are considered top of the line and capable of competing with desktop x86 CPUs and graphics cards.
The current M-series lineup is not entirely capable of beating high-end desktop hardware, at least in terms of raw performance. But Mac computers still have their niche; and given the technological milestones broken by their very first line of custom ARM SoCs paired with TSMC’s strong lead in the manufacturing marketplace, Apple’s success is already influencing trends across the PC market.
The future of the semiconductor industry:
Drivers of the semiconductor industry have faced challenges as a result of logistical factors and recent disruptions to the global supply chain, leading to chip shortages. TSMC may have beaten out Intel in adopting cutting-edge manufacturing equipment and overall facing fewer setbacks over the last few years, but its dominance is not firm, and Apple is not its only customer. Major chip designers Intel, AMD, NVIDIA, and Apple are all planning to release new generations of processors within the next couple of years. Intel has announced an aggressively optimistic roadmap through 2025 and is clearly ramping up to make a comeback. AMD is scheduled to release a chip based on its long-awaited Zen4 microarchitecture, which will be produced by TSMC using its 5nm process node.
Little is known about Apple’s next-generation M2 chip, which will replace the Apple M1 in the next generation of MacBook Air models. The M2 will not be as powerful as the M1 Max or Pro, but Apple is expected to introduce M2 Pro and Max chips for future generations of MacBook Pro models.
Image: Intel’s new node naming structure.
Source: Intel Renames 10nm ESF Node to 7nm, 7nm to 4nm, 5nm to 3nm in Roadmap Update.
The takeaway:
Companies specializing in electronics manufacturing and design must carefully consider to whom they trust the chip fabrication process, and their decision-making should involve extensive cost-benefit analysis. A plethora of potential partners exist to fill this role, and the past and present partnerships between semiconductor manufacturers and their high-value customers can often be indicative of strategies worth emulating.
While this article does not address the security aspect of the split hardware manufacturing model, “trust” in this sense is always something else to consider (What might the risks be of trusting 3rd-party manufacturers?). The key to developing a winning strategy will be to identify innovative semiconductor manufacturers to design custom solutions that meet the performance, energy, material, and security needs of products and their users. With this, PreScouter can help.
For semiconductor players, adopting and executing on the aggressive roadmaps for integrating advanced lithographic processes into production will likely be necessary to continue competing in this space. The ARM architecture’s rising popularity and shifting use cases ought to be considered, too. Further researching and analyzing the factors that lead Intel to fall behind within the chip manufacturing industry and others to rise to the ever-increasing demand will no doubt reveal patterns that will be useful in formulating strategies and roadmaps. PreScouter is here to help companies do just that.
