Flashback: how the Apple M1 evolved out of Apple’s iPad chipsets

The first custom Apple chipset, the Apple A4, launched in 2010 with the original iPad and was also featured in the iPhone 4 a few months later. The A4 was manufactured by Samsung and used an enhanced Cortex-A8 CPU core dubbed “Hummingbird”.

Hummingbird was co-developed by Samsung and Intrinsity and was announced in 2009 as “the world’s faster ARM Cortex-A8 processor”. Multiple customizations had to be made for the core to hit its 1GHz target. Apple acquired Intrinsity just months after it unveiled the iPad. And a couple of years before that it had acquired PA Semi.

After those key acquisitions, Apple set to work on in-house chipset designs to use in its portable products. Today’s story begins in 2012 as we will focus on the enhanced X-series of chips, which are predecessors to the groundbreaking Apple M1. The AX chips are used mainly in iPads, but they occasionally showed up in Apple TVs as well.

The second generation iPad introduced the Apple A5 to the world in 2011. It still used off-the-shelf components, Cortex-A9 CPU cores from ARM and PowerVR SGX543 GPU cores from Imagination. The third-gen iPad arrived a year later with an enhanced version of that chip, dubbed the Apple A5X, which got the ball rolling.

The A5X doubled the GPU cores (from MP2 to MP4) and also featured a new quad-channel memory controller, which offered data transfer speeds of up to 12.8GB/s, roughly triple the bandwidth of the A5.

Future AX chipsets would follow the same game plan – use the same hardware, just more of it. Tablets are larger than phones, meaning they have bigger batteries and more surface area to dissipate heat, so they could handle the more powerful chipsets.

The Apple A6 is notable for introducing the first custom CPU core designed in-house by Apple, called “Swift”. The GPU still came from Imagination. The A6X was a bit disappointing in that it only added an extra GPU core.

A couple of years later came the Apple A8X, the first in the series to expand the CPU hardware as well as the GPU. It added an extra Typhoon core, for a total of three, while the GPU core count was doubled to eight. The A9X went back to having the same CPU as the regular A9, but that was the last time – from then on all AX chipsets would have bigger CPUs

2016’s Apple A10 chipset was the first from the company to adopt a big.LITTLE architecture. It had two big Hurricane cores along with two small Zephyr cores. A year later the A10X came with three of each, while also doubling the GPU core count.

Little cores are great for efficiency, but having more than a few doesn’t add much performance. That is why the Apple A12X chipset from 2018 only doubled the big CPU core count (to four), while using the same number of small cores (also four). The GPU was upgraded to a 7-core design, an 8-core version would arrive in 2020 as the Apple A12Z.

Let’s jump to 2020 – after years of using Intel processors, Apple bid them farewell and announced the first batch of Apple M1-powered Macs. This also marked a transition away from x86 and towards ARM, the same ARM instruction set that powered its iPhones and iPads.

And that is no coincidence, the Apple M1 used slightly modified versions of the components in the A14 (the chip inside the iPhone 12 and 4th gen iPad Air) – the big Firestorm cores and the small Icestorm cores, the same GPU architecture as well.

But as we’ve already seen, the trick to making the chipset faster is to add more cores. The M1 doubled the big CPU cores and doubled the GPU (though it offered chips with 7-core GPUs as a cost-saving measure). Like with the 12X, the small CPU cores were left untouched. It helped that Apple’s designs were already in the lead in terms of both performance and efficiency (TSMC deserves some of the credit for that), so the M1 handled desktop tasks with ease, even when passively cooled.

The Apple M2 chipset that was announced earlier this month follows the same pattern, though this time it is based on the A15 chipset (iPhone 13). The M1 had Pro, Max and Ultra variants, the M2 certainly will as well.

These just use different multipliers, e.g. the M1 Pro has 50% or 100% more big CPU cores than the base M1 and double the GPU cores. The Pro cut the small cores down to two, but as already discussed only a few of those are necessary. The Max uses the same CPU formula, but offers 3 to 4 times as many GPU cores as the base M1. The Ultra doubles the CPU and GPU resources (it is actually constructed of two Pro chips).