Apple Silicon, Rosetta, M1, M2, SoC: Why these terms matter to every computer buyer | ZDNet

Apple Silicon, Rosetta, M1, M2, SoC… 

If you’re interested in Macs, you’ve probably heard one or more of these terms. But unless you’ve been closely following Apple’s technical progress, it might not be clear what they all mean and why they matter to your future buying decisions.

With Apple expected to make a new set of Mac announcements tomorrow, I want to answer a relatively common reader question: What does it all mean?

Understanding the CPU

A computer — whether Mac, PC, Linux machine, Raspberry Pi, or even the embedded brain in your microwave — consists of a set of components. Some manage input, getting data into the machine. Some manage output, presenting that information to you or doing a task (such as popping popcorn or displaying lifelike pictures in your video game). Some components store information, either temporarily or permanently. Some manage connecting to one or more networks (Ethernet, Bluetooth, Wi-Fi).

The processor, the central processing unit (or CPU), orchestrates all of these elements. The CPU processes sequences of instructions, does calculations, makes decisions, and tells data to move. In practice, modern computers have many processors, but some are used for special purpose calculations (like your GPU for graphics). But basically, at the center of it all is the CPU.

Also: Introducing my design for an Apple Silicon-based Mac Pro

Generally speaking, the faster the CPU, the faster the machine. That’s a wild over-simplification because even though a super-fast CPU might process instructions and calculations at warp speed, if it takes the system bus forever to move data around, it doesn’t matter. It’s like driving a Koenigsegg Agera RS in rush hour. The car may be one of the world’s fastest cars, but if the road is bottlenecked, it’s not going anywhere.

In reality, balance is important. If you’ve ever built a PC, you probably know that it doesn’t make sense to pair a high-end GPU with a mediocre CPU because the CPU will bog down the graphics. It doesn’t make sense to pair a super-fast CPU with generic, middle-of-the-road memory because the memory bus won’t be able to handle what the CPU throws at it.

This balance issue is, in fact, part of what has made Apple devices so successful. Because Apple (with a few limited exceptions) has always controlled the entire component mix of the machine, the company has been able to balance performance well. For fast machines, that means higher-end components. That means saving money on parts that won’t be fully utilized for slower machines.

Each CPU consists of one or more cores in almost all but the most simple machines. Cores are actually the processing units. A CPU with multiple cores has a traffic management aspect that manages the flow of instructions to each core.

Multiple cores can increase performance considerably for problems that can be split into parallel processes. Many modern processes work well in parallel, especially graphics, data crunching, AI, ML, and AR-related tasks.

Also: Introducing the Mac Mini Pro… again

CPU vs. SoC

Recall that I discussed that a computer consists of many components, most of which are integrated circuits (also called chips). Many computers also have many processors. I mentioned the CPU (the central processing unit) and the GPU (the graphics processor).

Some Macs include additional, special-purpose processors like the Neural Engine (for AI and machine learning) and a media engine (for non-game video encoding and decoding).

For many years, most powerful computers consisted of all the various system chips located as separate packages, soldered onto a motherboard. Some of the components, like memory and add-on boards, could be inserted via connectors on the board. This allowed for scalability and flexibility, but there are heat and propagation speed challenges with separate discrete packages soldered to a board. Many people who built their own PCs are familiar with this architecture.

The SoC (or system on a chip) architecture is different. Although the earliest SoC implementations were made for LED watches in the 1970s, they didn’t have processors in them. Instead, modern SoCs can be traced back to the 1990s. That’s when chip-making machines had evolved to where they could create components small enough to hold a whole system on one die.

Quick glossary minute: A die is the actual semi-conductive material that does the work. Most people think of a chip as technically a package that holds a die (sometimes more than one) and the wires to connect the die to the rest of the system. The chip is, in fact, the die inside.

Unlike the motherboard/CPU/memory model common inside most PCs, iPhones and the latest Macs are based on SoCs. These are chips that contain not only the processor (or processors) but memory, communications, even flash storage, and more — all inside a single package.

SoCs have enormous performance benefits because all the electrical impulses have to travel far shorter distances. Because the distance is shorter, the current used can be considerably less. And because the current is lower, there’s less power used. Less power used means less heat and longer battery life.

It wasn’t until SoCs could be successfully fabricated with tens of billions of transistors that they were suitable for high-performance personal computers. They can be now, so now they are.

Apple Silicon

By now, you’ve undoubtedly heard that Apple has stopped building Intel-based Macs and instead is building them based on something called Apple Silicon. Chips are made out of silicon (number 14 on the Periodic Table, symbol Si). So Apple Silicon is silicon chips from Apple. But what does that really mean?

Intel’s CPUs are often called x86 chips because they’re derived from the 8086 architecture and instruction set. The very first PCs were built using the 8088 and 8086 processors, way back in the late 1970s. Since then, there have been numerous generations (both by Intel and AMD), culminating in Alder Lake in 2021.

Also: Will Apple Silicon kill the Hackintosh? The odds against a self-built MacOS Arm computer

Apple’s Mac computer line has jumped chip families four separate times. The original Mac 128 introduced in 1984 used a 68000 processor. Apple stayed with the 68K family for 10 years, then jumped to the PowerPC processor in 1994. Apple stayed with PowerPC-based Macs for 12 years, then jumped to PC standard Intel chips in 2006. This was the dawn of the Hackintosh era because most Macs were generally compatible with standard PC architecture components.

Then, in 2020, Apple jumped to Apple Silicon, a chip architecture based on ARM, the processor type used in most smartphones, including iPhones. In fact, the A-series chips, like the A15 Bionic used in iPhone 13s, are essentially the same chip architecture as the Apple Silicon M1-based current-model Macs and iPad Pros.

For almost all activities, the Apple Silicon-based Macs blow away their equivalent Intel-based Macs in terms of performance, battery life, and heat. Apple’s Silicon-based Macs are an unqualified win for both Apple and consumers.

Apple, of course, doesn’t have to rely on Intel for its chips. To be fair, Apple still doesn’t make its own chips. It also doesn’t manufacture its own iPhones. The company relies on a vast supply chain to produce in the quantities it sells. But the company now designs its own chips for the Mac. By removing its reliance on Intel’s innovation teams, Apple can engineer to its specific priorities. The result is apparent.

But the Apple Silicon transition was a win for Apple and consumers solely because of something called Rosetta 2.

Rosetta 2

Intel x86 and Apple Silicon ARM are two vastly different architectures. That means they process data very differently, and they use different instructions to do so.

All major applications are built using computer code. Programmers write code in a higher-level language, basically a language that humans can read, debug, and maintain. But that human-readable language is highly inefficient to a processor, so the code is converted into machine code.

Machine code lacks the niceties of higher-level languages that make it readable and maintainable by humans. Still, processor chips understand it and can execute it far faster than human-readable code. When you install a program on your computer, you’re usually installing compiled code — code that’s been translated to be read by the processor, not by you or a programmer.

Until 2020, most Mac apps were distributed with code compiled for Intel processors. If that code was handed to an Apple Silicon-based Mac, the M1 processor wouldn’t be able to do anything with it. The computer would respond with the machine’s version of “Eh?”

The difference in architecture is profound. The languages and even the structure of the languages understood by Intel processors and Apple Silicon processors are wildly different. This was also the case going from 68000 to PowerPC and from PowerPC to Intel.

In other words, Apple has solved the processors-don’t-speak-each-others’-languages problem a number of times before. They do so using a combination of translation and emulation.

With new Apple Silicon machines, this is done by Rosetta 2. When you try to open an Intel binary on an M1-based machine, MacOS hands it off to a program called Rosetta 2. Rosetta 2 does an on-the-fly translation of the x86 code into Apple Silicon code, saves it, and then runs the translated code. Some elements are emulated while other elements are completely transcoded.

The first time you run an Intel-based program, it may take a little while for the program to start running. That’s because Rosetta is doing a translation pass first. Subsequent runs will then be faster because the translation has already been done.

The first time you run your first Intel program on an M1-based Mac, MacOS may ask you if you’d like to install Rosetta 2. This is a definite yes because that opens the door to all those Intel apps that you might already have. MacOS will pull down the Rosetta 2 code from Apple, install it on your computer, and become able to run Intel-based applications. (Historical note: Rosetta was originally used to translate from PowerPC to Intel more than a decade ago. That’s why we’re at Rosetta 2.)

After being translated with Rosetta, some Intel-based applications will actually run faster on Apple Silicon than they used to on your Intel-based Mac.

Let’s take a minute to talk about developers. Developers code using higher-level languages, but they have to compile their project for each architecture. Since 2006, most developers have been compiling for Intel-based machines. When Apple introduced the M1, the company made the process of compiling that same code to Apple Silicon relatively easy. But it’s still a non-trivial investment in developer time to create the new version.

While most major developers have made the jump, others (either because it’s too big an investment, they have other priorities, or they don’t see a good business reason) are still shipping Intel-only apps. Some applications that users rely on are older, not maintained, and will never be updated for Apple Silicon.

That’s why the Rosetta 2 translation/emulation capability was so critical to consumer acceptance of the new Apple Silicon architecture. When the M1 first came out, developers were understandably skeptical of how well it would do compared to Intel. They reasoned that if Intel apps wouldn’t run, or ran very poorly, that users (who are pretty reliant on the software they are used to) wouldn’t buy the new Apple Silicon-based machines.

But now that Apple Silicon has been out for a year or so and has been a smashing success, more developers are investing the time to create Apple Silicon-native apps.

It’s a win for both developers and users when code is compiled for Apple Silicon. In general, native Apple Silicon programs run faster than Rosetta-translated ones. So if a developer recompiles their apps for Apple Silicon, they get an almost automatic performance boost.

Also: Sluggish apps on your M1 Mac? Check this first for a possible fix

I found, for example, that Rosetta-translated Chrome was okay but a bit sluggish. When I replaced that with the Apple Silicon version of Chrome, it was far faster. For help finding out which of your applications are Native, Intel, or “Universal” (meaning the application contains native code for both Intel and Apple Silicon), read this article. It will show you how to find what is running what.

M1, M1 Pro, M1 Max, and M2

M1, M1 Pro, and M1 Max are processor models that Apple uses inside its computers. As of this writing, the M2 is the name pundits have assigned the expected next generation of Apple Silicon processors, but it doesn’t exist yet.

Also: Inside the new Mac mini: Does Apple’s M1 architecture really leave Intel behind?

The M1 is the original Apple Silicon processor for Macs, announced in 2020. The M1 Pro and Max were announced in the fall of 2021. All M1 processors have four different types of cores: performance cores (power computing), efficiency cores (slower, for more pedestrian work, but take less power), GPU (for graphics performance), and Neural Engine (for AI and machine learning).

The difference between the models is in how many cores. The M1 has four performance cores and four efficiency cores. The M1 Pro and M1 Max have only two efficiency cores, but either 8 or 10 performance cores (depending on how much you want to spend). All three have 16 neural cores. The biggest difference is in GPU cores. The base M1 has 7 or 8. The M1 Pro has 14 or 16. And the M1 Max has 24 or 32. Adding it up, the max Max comes with 60 cores!

But these are SoCs, so memory and storage come right on the chip. Depending on the chip, you can go from 8GB RAM up to 64GB RAM and 256GB of flash storage up to a whopping 8TB on the 16-inch MacBook Pro with M1 Max. That additional storage will cost you, to the tune of a few thousand bucks more. But if you need it, you need it.

The size of the chips are considerably different. The M1 is about a quarter the size of the M1 Max. This makes sense because the M1 Max crams a lot more onto its die. It will get smaller when the die production process is able to use smaller components — technologies that the chip makers are working on actively. But for now, it’s just interesting to note that the M1 has 16 billion transistors, while the M1 Max has 57 billion transistors — all in the space of about the size of a fingernail.

Final thoughts

M1, M1 Pro, M1 Max, and the expected upcoming M2 are all Apple Silicon. Each just has a little more oomph than the next. When you look at purchasing your next Mac, you’ll want to do a typical price/performance analysis to decide which of those SoC chips meets your needs and your budget.

Now that you know more about what’s going on under the hood, are you planning to move to Apple Silicon? Have you already done so? If so, what have your experiences been with translating Intel apps to Apple Silicon with Rosetta 2? Did you even know that was happening? Do you have your eye on a sweet M1 Max? Are you holding out for the fabled M2? Let us know in the comments below.

You can follow my day-to-day project updates on social media. Be sure to follow me on Twitter at @DavidGewirtz, on Facebook at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, and on YouTube at YouTube.com/DavidGewirtzTV.