an ARM in a Mac?


Rumours of Apple dropping Intel in favour of an in-house ARM-based processor in Macs are not new.  They have been floating around for years.  While, to date, all have proved premature, Apple’s design prowess shown in its A-series SoCs suggests this goal is within reach.

The other week, Mark Gurman at Bloomberg renewed this thread, suggesting ARM-based Macs will appear in 2021.  Details included the project codename, the number and type of CPU cores and the presence of an integrated GPU, where the CPU and GPU will be based on those in the upcoming A14.  OK, that all sounds good, but Apple has much more up its sleeve.  It’s time to think about what else they could bring to the table, so to speak.

CPU and GPU

Let’s assume that the Bloomberg article is correct in that the CPU and GPU designs of the A14 will be the basis of an Apple-designed processor for the Mac.  I will call them the M-series processors.  I also agree there will be various configurations and numbers of low and high-power CPU cores.  That is all fine.  But, that is not the end of the story.  The A-series processors are so-called System on Chips (SoCs), having circuits beyond the CPU and GPU to perform specific computational tasks and  housekeeping functions.  Over the years I wrote that these groups of circuits, or “blocks”, outside of the GPU and CPU are important and need to be part of the discussion.  A number of these blocks provide for specific functionality or performance.  Some of these have already made their way to the Mac.

T1 & T2

The T1 and newer T2 are Apple-designed ICs in Macs that run some form of iOS, with one early account suggesting the T1 ran a version of watch-OS.  Blocks in the T1 and T2 are from the A-series processors.  The late 2016 MacBook Pro with TouchBar was the first Mac to include the T1.  It ran the TouchBar’s graphics, the fingerprint sensor and enhanced device security, to name three functions.  The T2 is now present in pretty much every Mac, with the lone standouts being some iMacs that have not seen a refresh in a while.

The T2 has evolved into an important component in the Mac line.  It includes a Secure Enclave (SE) for the fingerprint sensor, as found on all laptops, and device security used in all instances, including the Mac Pro.  It would have some image processing, such as a GPU and, maybe, hardware accelerators for the Touch Bar and video processing.  This latter functionality is put to use in video processing in the Mac mini, for example.  Finally, the T2 would also include and ECC a.k.a. storage controller.  Apple has been designing its own memory controller for storage for many years now.  This is a critical block of the T2.  All of these blocks of the T2 were originally developed for and used in the A-series processors.  It is pretty straightforward to see them being integrated on any M-series processor destined for the Mac.

Moving the critical blocks of the T2 to an M-series processor has many benefits.  Right off the bat there is one less IC on the motherboard, reducing both costs and space requirements.  Further, one would only be communicating within the M-series processor saving time and energy, compared to communicating with an external chipset.  The inclusion of blocks from the T2 is a given.

the NE wildcard

Will a Neural Engine (NE) be included on any M-series processor?  This is a very interesting question.  The NE is a large block on the A-series, performing machine learning and AI functions.  Apple’s NE has only been around since the A11, but it appears to be integral to the current functionality and workload.  In particular it is central to processing related to the iPhone’s camera. Quite a bit of time in the September 2019 Keynote was devoted to this task.  It was also noted that the CPU, GPU and NE are all involved, where a dedicated block balances computational loading between them.

We are back to the original question: will the NE be included in any M-series processor?  We know it is important for the camera, but Macs do not have this functionality.  Yes, there is a webcam in all Mac laptops, and iMacs, but this is nowhere near the same.  A webcam is simple video capture for communications.  This would imply the NE is not needed.  Is there another use for it?  Could it be used, again with load balancing, to finally turn Macs into gaming machines?  Could it work in co-ordination with the GPU and CPU for image processing and graphics rendering?

It might also be used for machine learning and AI processing.  I do not see this as a big requirement on laptops and desktops at the moment, but that does not imply it won’t be in there.  Is this because of a lack of general applications for this type of computing.  Including a dedicated block i.e. the NE could enable a new set of functionalities.  It is, at the very least, interesting to ponder.

If either the CPU or GPU were not up to snuff for the M-series discussion would be dead in its tracks.  I don’t think this is the case.  I think these two processing units are up to the task.  If we still need a Rosetta-style interface that too could be hardwired as a block of the M-series.  In 2006 it ran as a separate bit of software.  In an Apple-designed 5nm M-series processor it could be onboard.  The rest of of the blocks of any M-series processor will though prove to be quite interesting.  They might just be the bits that differentiate future Macs from the Intel contemporaries.  As I have said many times, don’t forget “other”.  Time will tell.