ARM64

Every processor only understands a certain set of very simple instructions: add two numbers, fetch something from memory, jump to a different point in the program. This list of permitted instructions is called an instruction set. ARM64 is one such instruction set, developed by the British company Arm. The 64 stands for the fact that the processor works with 64-bit long numbers and memory addresses, meaning it can handle very large values and a lot of working memory. The most important competitor is called x86-64 and comes from Intel and AMD. A program compiled for one instruction set does not run on the other without further ado.

Why ARM migrated from the phone into the laptop

ARM chips were originally designed for devices that have to get by on a battery. Efficiency was therefore the most important goal from the very beginning. This very characteristic has since become interesting for other devices as well. A notebook with an ARM64 chip often runs a whole workday without a power outlet, and sometimes doesn’t even need a fan.

For data centers the same advantage counts, just on a different scale. There, tens of thousands of servers are running, and power consumption is a significant cost factor. If a chip draws a third less energy at the same performance, the operator saves millions. Amazon, Google, and Microsoft are therefore developing their own ARM64 server chips instead of just buying them.

On top of that comes a business model that differs from Intel’s. Arm doesn’t build chips itself, but sells blueprints and licenses. Anyone who wants to can build on them and design their own processor. Apple has done this with its M-series chips, and the success of these chips has set the entire industry in motion.

Few instructions, many cores

ARM64 belongs to the so-called RISC family. The idea behind it: the processor masters relatively few, but uniform instructions. All instructions have the same length, which greatly simplifies decoding inside the chip. x86-64, by contrast, has accumulated ever more special instructions over decades, which vary in length. There, the chip needs more circuitry just to figure out where an instruction even ends.

Less overhead per instruction means less die area and less power. The space freed up can be used for additional computing cores. ARM64 server chips therefore often have 64, 128, or even more cores, while classic server processors get by with considerably fewer. For tasks that can be split up well, such as serving web pages, that’s a clear advantage.

The biggest practical problem is software. Programs exist as finished, compiled machine code and each only fits one instruction set. That’s why there are translation layers like Apple’s Rosetta 2, which convert x86 programs for ARM64 on the fly. This works surprisingly well, but costs speed. A common misconception, by the way, is that ARM64 is automatically faster than x86-64. More efficient, yes, faster only depending on the chip and the task.

ARM64 in devices you use

If you’re holding a smartphone in your hand, you’re holding an ARM64 processor. That’s true for practically all Android devices and for every iPhone. Tablets, smartwatches, TVs, and the Raspberry Pi also work with this instruction set. Since 2020, this has additionally applied to all Macs that Apple sells.

In the news you usually encounter ARM64 under a product name. Apple M4, Qualcomm Snapdragon, Amazon Graviton, or Nvidia Grace are all ARM64 chips. When you download software, you sometimes have to choose between versions for arm64 and for x86-64. This label only tells you which instruction set the program was compiled for.

For the stock market, Arm is a topic of its own. The company went public in 2023 and is valued even though it doesn’t manufacture any chips itself. The value lies in the licenses and in the fact that the architecture now also plays a role in the AI space. In many AI servers, an ARM64 processor sits alongside the actual graphics chips and controls their work.