The Apple M2 is not a new chip
The M2 is complex and expensive — will it do well?
Apple just announced the new M2 chip at the WWDC 2022, the second iteration of the M-series that has derived from the famous iPhone CPUs. Over the past years, Apple has consistently had not only the best smart phone CPU but also beat all other competitors in GPU areas as well as overall efficiency. This shows us how good their design is — but will it be sustainable enough to keep the performance crown in the coming years?
With the advent of that CPU design in the mac computer field, Apple was able to sell portable computers with extremely efficient processors yielding in excellent performance paired with a fantastic battery life. As I wrote elsewhere, the real benefit of the M1 is not its sheer power but its efficiency. Something it has received in the cradle because it descended from the iPhone where Apple has years long experience with creating efficient processors.
Whats new
The next step is the M2, built using second-generation 5-nanometer technology. It features an incredible 20 billion transistors on an even larger die resulting in increased manufacturing costs. Yet the raw single core performance increase is a mere 18% as apple stated. The bigger gains come from the GPU side and the neural engine which is also used by video transcoding tasks.
The M2’s performance cores are now based on the A15’s Avalanche core with some modifications to better handle larger memory configurations (or take advantage of) — following how the M1 used modified Firestorm cores.
One advantage of the new core pair Firestorm/Blizzard over the Firestorm/Icestorm used in the M1 and the A14 is the increased efficiency of the low power cores, as Anandtech reports. This is an advantage you will not directly notice, but which will lead to better battery life which is a little more important in the mobile phone area than it is for computers.
The problem
One of the “problems” of the M1 was its memory performance. But wait, the M1 has an excellent memory performance!?
Yes it has. Its Universal memory architecture was a game changer for the Macs that had the chip. Taking CPUs this architecture allowed for extremely fast memory accesses for all processor cores.
So whats the problem?
The problem is that this SoC also has a huge GPU included that relies on that same memory too. Speaking GPU, a memory bandwidth of around 60GB/s is nothing nVidia or AMD would be interested in given their GPUs memory bandwidths are in the range of several hundert GB/s.
So yes, that combined SoC has a problem: It cannot offer the memory bandwidth all these CPU and GPU cores would be able to request essentially leading to some sort of throttling. Besides the additional cores, much of the gained performance in the GPU area might come from the twice as fast memory interface the M2 has compared to the M1. We will see to what extend this also allowed the yet unannounced M2 Pro and Max will be able to benefit from that.
The future
Because in the future an increasing amount of applications will be relying on graphics tasks and AI, the move to increase the performance of these blocks seems wise. And they help the CPU to be more efficient by allowing more tasks being offloaded to them leaving the CPU doing more “basis” tasks.
So while most of the increased raw CPU performance of the M2 may just come from increased clock speed and larger caches, the more powerful memory interface supporting LPDDR 5 and the added complexity of the GPU and Neural Engine make this new chip a good basis for the coming Pro and Max editions.
