In my last post I mentioned that Apple’s M1 chip is amazing. I mentioned that it was getting terrific performance and stunning battery life. I mentioned that this is going to have an industry-wide impact over the coming years. Maybe the coming decade. Yes, I mentioned that Apple did all those things. But I didn’t say how they did it.
First, the M1 is a “system on a chip,” or SoC. A typical PC or laptop has a number of discrete components to it. There’s the microprocessor, the memory, maybe a graphics chip, and other doodads. These various parts are all wired up to each other on the same circuit board. An SoC is different. It takes all those components and puts them on the same chip. Doing this gives you a lot more efficiency.
Imagine you were ordering off a menu in a fancy restaurant. The menu is filled with international dishes from all over the world–and they are each listed in the languages of those places. Fortunately, you have a friend who speaks Italian, another one who speaks French, and a third one who speaks Japanese. To navigate the menu and order your meal, all you have to do is call each of them in turn when you encounter their language on the menu. They would give you the information you need (an English translation) and you’d be good to go. It’s not very efficient, but it gets the job done.
Now imagine that your three friends are at the table with you. That’s an SoC. Having those resources closer to you is way more efficient. It’s true for you and your multi-lingual friends, and it’s true for a CPU and all the other components it needs to talk to. Everyone separate is inefficient. Everyone at the same table (or on the same chip) is way more efficient. In the case of Apple’s M1, they’ve put the graphics processors, the security, the encryption, the I/O, the storage controller, and the machine learning all into one chip with the CPU. They all use the same pool of memory, too. Yep. The memory is on the M1 chip, too.
That’s how the M1 gets things done more quickly and uses less energy to do them. But there’s more. That CPU is divided up into several “cores.” Many computing tasks can be divvied up between multiple cores working simultaneously, so the more of them you have, the faster the work gets done. The clever thing Apple has done here is that in addition to four “performance” cores, they’ve also included four “efficiency” cores. These latter handle the small stuff, the easy stuff, so that the big boys can focus on the big jobs. The efficiency cores use less electricity than the performance ones do, which saves battery. It also means you have less heat to dissipate, so little to no active cooling is required.
The M1 is beating the pants off other chips that use 45, 65, or even more watts. And it uses a maximum of only 15 watts to do it. That’s why my new MacBook Air will get six more hours of battery life than the previous Air did, even though it has the exact same battery as the older model.
We haven’t even discussed the fact that the M1 CPU uses an ARM instruction set which is simpler than that used by x86. ARM chips have been powering mobile devices for decades. And Apple has been making their own ARM chips since 2012, so they’ve had some time to figure out how to eke every last bit of power and efficiency from them.
Finally, consider the efficiencies you can achieve when you create the operating system, the CPU and the design of the whole computer. It’s quite common to have a laptop made by Dell using a CPU made by Intel running an operating system made by Microsoft. Dell does not control design and engineering choices made by Intel any more than they can control the choices made by Microsoft. Because Apple makes the whole stack themselves, the software, the hardware, and the CPU are each bespoke designed to work with one another.
Add all this up and there’s your miracle. That is how they did it.
Oh, one more thing. The M1 is also created using a 5 nanometer process so it can fit 16 billion transistors on the chip. This is probably the easiest thing that Intel and AMD can do to catch up. I expect later on this year AMD will have their own 5nm chips coming to market. Intel, I dunno. They can’t seem to get smaller than 10nm, so who knows with them.