Discussion Apple Silicon SoC thread

[Eug]

M1
5 nm
Unified memory architecture - LP-DDR4
16 billion transistors

8-core CPU

4 high-performance cores
192 KB instruction cache
128 KB data cache
Shared 12 MB L2 cache

4 high-efficiency cores
128 KB instruction cache
64 KB data cache
Shared 4 MB L2 cache
(Apple claims the 4 high-effiency cores alone perform like a dual-core Intel MacBook Air)

8-core iGPU (but there is a 7-core variant, likely with one inactive core)
128 execution units
Up to 24576 concurrent threads
2.6 Teraflops
82 Gigatexels/s
41 gigapixels/s

16-core neural engine
Secure Enclave
USB 4

Products:
$999 ($899 edu) 13" MacBook Air (fanless) - 18 hour video playback battery life
$699 Mac mini (with fan)
$1299 ($1199 edu) 13" MacBook Pro (with fan) - 20 hour video playback battery life

Memory options 8 GB and 16 GB. No 32 GB option (unless you go Intel).

It should be noted that the M1 chip in these three Macs is the same (aside from GPU core number). Basically, Apple is taking the same approach which these chips as they do the iPhones and iPads. Just one SKU (excluding the X variants), which is the same across all iDevices (aside from maybe slight clock speed differences occasionally).

EDIT:



M1 Pro 8-core CPU (6+2), 14-core GPU
M1 Pro 10-core CPU (8+2), 14-core GPU
M1 Pro 10-core CPU (8+2), 16-core GPU
M1 Max 10-core CPU (8+2), 24-core GPU
M1 Max 10-core CPU (8+2), 32-core GPU

M1 Pro and M1 Max discussion here:

Page 78 - Discussion - Apple Silicon SoC thread

Page 78 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

M1 Ultra discussion here:

Page 109 - Discussion - Apple Silicon SoC thread

Page 109 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

M2 discussion here:

Page 127 - Discussion - Apple Silicon SoC thread

Page 127 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Second Generation 5 nm
Unified memory architecture - LPDDR5, up to 24 GB and 100 GB/s
20 billion transistors

8-core CPU

4 high-performance cores
192 KB instruction cache
128 KB data cache
Shared 16 MB L2 cache

4 high-efficiency cores
128 KB instruction cache
64 KB data cache
Shared 4 MB L2 cache

10-core iGPU (but there is an 8-core variant)
3.6 Teraflops

16-core neural engine
Secure Enclave
USB 4

Hardware acceleration for 8K h.264, h.264, ProRes

M3 Family discussion here:

Page 215 - Discussion - Apple Silicon SoC thread

Page 215 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

M4 Family discussion here:

Page 263 - Discussion - Apple Silicon SoC thread

Page 263 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[Eug]

Why would an app need 16GB of memory, virtual or otherwise? What are they expecting people to do on their iPads???

Creative professional applications like Adobe Illustrator, etc.

 

[SpudLobby]

Yeah, no way it will be annual. Cadence will be driven by volume.

iPhone gets annual, since it has Massive Volume, and is the most important Apple product.

Next biggest Volume is the Entry Level Mac/iPad Pro chip. Cadence for iPad Pro was around 18m-2 year and I expect they will just keep that.

Pro-Max-Ultra - big drop in volume here so firmly in the 2 year or longer cadence.

IMO, the chips are so powerful that 2 year cadence is fine.

This is completely correct IMO, pretty much my views, I'd set 18 months as the lower bound for stuff like the entry-level Macs. It doesn't make sense to do annual refreshes for the massive SoC's, it only somewhat makes sense for the M-series given the profitability, volume of the iPhone and yield/die size relationship.

Apple's economics (their profitability, reuse of supply chain components) allow them to spend on transistors and leading process nodes, to be sure, but the idea that they were going to e.g. have the M3 Pro/Max/Studio/Pro by 2023 is just really absurd. You're going to see a lot of weird mismatches, I think, between the lineups due to the economics here, e.g. maybe an M2 Pro/Max comes Q4 2023 while we see an M3 chip in spring 2024, or they may even skip the M2 series for those larger packages.

Kind of enjoying the people losing their minds over this though. It was pretty obvious the M2 would be an A15 extension like the M1 was of A14, I think. Hell, the M1 Pro/Max/Studio coming almost a year or more after the M1 was an omen of sorts too, and there were no ST changes, still Firestorm @ 3.2GHz.

 

[moinmoin]

It may well be the case that Apple consolidates the best selling iPad and MacBook models around the base M chip which then represent enough quantity to get yearly refreshes along with the A series. Agree that that's hard to see being extended to the bigger M Pro/Max/Ultra chips though.

 

[poke01]

M2 is also used in the 13" MBP. Calling any laptop a "Pro" machine whilst only allowing a single external display is an insult.

The 13" MBP is not "Pro" at all. The real Pro laptops from Apple are the 14" and 16" MBP.

 

[Eug]

It may well be the case that Apple consolidates the best selling iPad and MacBook models around the base M chip which then represent enough quantity to get yearly refreshes along with the A series. Agree that that's hard to see being extended to the bigger M Pro/Max/Ultra chips though.

Given the price point and target customers, I’d think the base iPad (which is the best selling model) will stay A series. Hell, the base iPad is already roughly as fast as my 2017 Core i5-7600 iMac.

 

[mikegg]

This doesn’t sound right. I’d be very surprised if M series is yearly.

Why not?

If it isn't yearly, why would they base the M2 on A15? Why not skip it and go straight to the A16 if it's not yearly? Why were there extremely confident reports from Mark Gurman that Apple planned to release the M2 in the old 13" MBP chassis in the Spring Event? While the M2 13" MBP didn't show up until WWDC, it clearly shows that the M2 chip was ready much earlier than WWDC and was only held up by the new Air design.

Also, the base M chip goes into iPad Air, iPad Pro 13, iPad Pro 11, Mac Mini, iMac, Macbook Air, 13" Macbook Pro. And they might put it in future devices like Apple TV/Console, AR headset, Macbook Air 15", etc.

That's a lot of devices from a single chip.

It makes a ton of sense to do annual updates to the base M at least.

It may or may not make sense for Pro/Max. It doesn't make sense for Ultra, I agree.

 

[Roland00Address]

Just a reminder people are "assuming" things go well with a yearly cadence, that a year is always 12 months and you always hit your marks.

In reality things can not be +2 / -2 months late / early, let alone +4 / -4 months.

With only 2 data points for the M1 generation, though we have 9 data points with the A series and 8 data points with the 64-bit. But also with the iPad X and Z series we know that cadence is not as nice as neat as the Apple iPhone A series.

 

[JasonLD]

Why not?

If it isn't yearly, why would they base the M2 on A15? Why not skip it and go straight to the A16 if it's not yearly? Why were there extremely confident reports from Mark Gurman that Apple planned to release the M2 in the old 13" MBP chassis in the Spring Event? While the M2 13" MBP didn't show up until WWDC, it clearly shows that the M2 chip was ready much earlier than WWDC and was only held up by the new Air design.

Also, the base M chip goes into iPad Air, iPad Pro 13, iPad Pro 11, Mac Mini, iMac, Macbook Air, 13" Macbook Pro. And they might put it in future devices like Apple TV/Console, AR headset, Macbook Air 15", etc.

That's a lot of devices from a single chip.

It makes a ton of sense to do annual updates to the base M at least.

It may or may not make sense for Pro/Max. It doesn't make sense for Ultra, I agree.

I don't see why Apple would need to be updating M series annually when none of the Apple product other than iPhone is getting yearly updates. 18-month cycle is the best bet for M series.

 

[Doug S]

Why would an app need 16GB of memory, virtual or otherwise? What are they expecting people to do on their iPads???

The iPad Pro already comes with 16GB of RAM in the models with 1TB or 2TB of storage, so all they are really doing is allowing the lower spec iPad Pros to approach the capability of higher spec one.

App developers are able to collect data on how much memory people are actually using, so maybe they were the ones who asked for this based on usage patterns they were seeing with some of their customers.

 

[mikegg]

I don't see why Apple would need to be updating M series annually

To sell more products. Base M goes into 7 Apple products right now.

 

[JasonLD]

To sell more products. Base M goes into 7 Apple products right now.

It is going to take more than a year to get M2 upgrades for all 7 Apple products you mentioned lol. And Apple won't bother going to M3 before that upgrade cycle is complete.

 

[mikegg]

It is going to take more than a year to get M2 upgrades for all 7 Apple products you mentioned lol. And Apple won't bother going to M3 before that upgrade cycle is complete.

Some products get the M2 first, namely the Macbook Air and the iPad Pros. The others follow. Even if the base M gets updated every 2 years, it's not going to go into all 7 products at the same time.

The point is that the base M is extremely important for 7 Apple products. It deserves to be updated annually. And it will be.

The fact that the M2 is based on A15 and that it was reportedly ready by Spring (March 2022) shows that Apple clearly has intentions to update it on an annual basis.

And it really isn't that much different from the A15 except for a few more cores.

 

[Heartbreaker]

Why not?

If it isn't yearly, why would they base the M2 on A15? Why not skip it and go straight to the A16 if it's not yearly?

It already isn't yearly. M1 Nov 2020, M2 June 2022. That's 19 Months.

iPhone gets yearly because they sell hundreds of millions of them yearly.

Macs selling about 1/10th of iPhones, and that's all Macs combined, and they need different chips to do that.

A more appropriate example is iPad Pro "X" chips averaged about 18 months sometimes one year, sometimes 2 years.

There is simply no need to update Macs every year, and Apple has historically had some long intervals between Mac updates.

They will do updates, when they feel their is a real benefit to doing updates. NOT updates for the sake of updates or following some arbitrary schedule.

I wouldn't be surprised if they also skip part of the lineup. The might do and M2 base chip, but then NOT do a M2 Pro/Max, but later do M3 Pro/Max without doing and M3 Base...

There are no hard and fast rules. Updates will merely be at Apple's convenience. And Apple's convienience will not lead to annual updates.

 

[Eug]

It already isn't yearly. M1 Nov 2020, M2 June 2022. That's 19 Months.

Actually, it's closer to 20 months. M1 was available November 2020 in customers' hands. M2 will be available in customers' hands in July 2022.

If they are able to update the base M as frequently as even every 18 months, I will be impressed. It's not going to be every year, and as @guidryp says, Apple is already telling us it's not going to be every year, by example with their real world releases.

I also think Mac Pro will be updated even less often, and it won't necessarily follow the M schedule of hardware updates, which is another reason I think they could possibly consider using non-M nomenclature. Then again, I suppose they could use SoCs made up of quad M2 Max / dual M2 Ultra (but not quad M1 Max / dual M1 Ultra).

 

[trivik12]

Considering Ipads are not updated yearly, I doubt Macs will be updated yearly as well. i guess volume is not there unlike iphone which sells > 10x than Mac/Ipads.

 

[Mopetar]

18 months doesn't hurt Apple all that much. They have enough different products that they can swap around releases so that they always have a new something for their usual events. They could easily do this by swapping between consumer/pro products for their event announcements.

It seems like we'll still see the cycle for iDevices run on a yearly refresh. Too many people still operate on the 2-year contract from carriers for Apple not to target a yearly refresh cycle.

 

[Doug S]

It is going to take more than a year to get M2 upgrades for all 7 Apple products you mentioned lol. And Apple won't bother going to M3 before that upgrade cycle is complete.

It shouldn't take long assuming the M2 Pro/Max tapeout happens around the same time as the M2 tapeout. Other than the die to die I/Os, it is all the same blocks just more of them. M2 Pro & Max require some extra steps for binning on number of CPU & GPU cores, and maybe they want to do a bit more verification since they go in higher end products. There would absolutely be more verification and more manufacturing steps in general for M2 Ultra & M2 "Extreme" (i.e. Mac Pro) due to the way they are made.

So it makes sense the Pro/Max/Ultra/Extreme come later, but there is no reason that has to take a year. It COULD take a year or even longer if Apple decides to stretch it out for their own reasons, or due to product design cycles for the Mac hardware that goes around Apple Silicon. But if the two designs taped out around the same time, let's say with the bigger one a month later, there's no reason they couldn't be shipping Mac Pros with the M2 Extreme six months after the first M2 Mac ships in July. Not saying that will happen, but there is no reason why it can't if that's what Apple wants to do.

Since the Apple Silicon transition isn't even complete yet we can't look at how long cycles are taking and assume that's how long they will take in the future. Nor can we look at Apple's product cycles with x86 and assume they plan to do the same thing with Apple Silicon.

 

[Roland00Address]

So question here, does anyone want a M1 iPad Mini (8.3” 3 by 2 display, largest iPhone compared below.)

I do not need a M1 I think an A15 is plenty, but I would appreciate extra ram and all the multitasking features they promised with the future M1 iPad OS update. Do we know if any of those features specifically need a M1 with things like gpu, ram, neural engine, etc or are we not going to get this stuff due to artificial segmentation?

 

[Eug]

So question here, does anyone want a M1 iPad Mini (8.3” 3 by 2 display, largest iPhone compared below.)

I do not need a M1 I think an A15 is plenty, but I would appreciate extra ram and all the multitasking features they promised with the future M1 iPad OS update. Do we know if any of those features specifically need a M1 with things like gpu, ram, neural engine, etc or are we not going to get this stuff due to artificial segmentation?

The rumour mill is saying there may be a 14" iPad Pro.

If true, what I take this to mean is a product line something like this:

12-13" and 14" iPad Pro (M series)

11" iPad Air (M series)

10.x" iPad (A series)

8.x" iPad Mini (A series)

 

[repoman27]

Last night I pulled together a bunch of data just to sanity check some of the assumptions here. I figured I'd share, because some items, like the AX series release cadence, were far less consistent than I had remembered. On the other hand, the average interval between releases ends up being 17.6 months, which is pretty much what everyone thought.

SoC	launch date	months between launches	notes
A5X	2012-Mar-16
A6X	2012-Nov-02	7.6
A7X	skipped		28nm node skipped
A8X	2014-Oct-16	23.4
A9X	2015-Sep-09	10.8
A10X Fusion	2017-Jun-16	21.2	10nm lead product
A11X Bionic	skipped
A12X Bionic	2018-Oct-30	16.5	15.9 mo. avg. for AX series
A12Z Bionic	2020-Mar-18		fully enabled A12X
A13X Bionic	skipped	skipped	N7P node skipped
M1 (A14X Bionic)	2020-Nov-17	24.6
M2 (A15X Bionic)	2022-Jul-01	19.4	17.6 mo. avg. for AX incl. M

The regular A series is obviously far more consistent, generally hitting the annual release window like clockwork. I thought Apple might do something along the lines of:

A = 12 months​

M = 18 months​

M Pro/Max/Ultra = 24 months​

M Mega = 30 months​

So maybe we see M2 iPads announced in September along with the A16 iPhones, and in October Apple announces the all new M2 Mac mini and refreshed 24-inch iMacs, along with the M2 Mega Mac Pro to conclude the Apple silicon transition. The gap to the M3 might be less than 18 months, and then the M3 Pro/Max/Ultra would follow that several months later. I could see Apple choosing to sit out N3 for the larger chips, depending on how initial yields look on that node.

Here are some shipment numbers for reference:

2021 shipments	All PCs	Mac	iPad
IDC	348.8	27.775	57.8
Gartner	339.769	25.983
Canalys	341.053	28.958	61.043

And here is an additional breakdown based on quarterly shipments from Canalys and x86 market share from Mercury Research:

Segment	2021 unit shipments
PC	341.053
Notebook	275
Desktop	66
Intel PC	250.317
Intel Notebook	200.372
Intel Desktop	49.945
AMD PC	61.852
AMD Notebook	51.394
AMD Desktop	10.458
Apple Mac	28.958
Apple iPad	61.043
Apple M Series iPad (estimate)	21.365
Apple M Series Total	50.323

 

[Roland00Address]

The rumour mill is saying there may be a 14" iPad Pro.

If true, what I take this to mean is a product line something like this:

12-13" and 14" iPad Pro (M series)

11" iPad Air (M series)

10.x" iPad (A series)

8.x" iPad Mini (A series)

Yeah I expect the same, but my fear is Apple will artificially segment features from the A series that is $500 in price, and the M series which the cheapest model is currently $600 in price. What I am saying is I understand sometimes you may need to segment features by real hardware (memory bandwidth, total ram, gpu, neural engine etc.) I just hope that the iPad Mini will suffer for it is in the liminal space where it is not a phone, and it is not a big iPad with better cooling and thus can handle a M1.

Is this fear unfounded, is this just me being on edge for I like the hardware and have desires? Well I want feedback outside of myself but also not from Apple Directly 🤣 from fellow nerd like people.

 

[Heartbreaker]

I figured I'd share, because some items, like the AX series release cadence, were far less consistent than I had remembered.
...
The regular A series is obviously far more consistent, generally hitting the annual release window like clockwork. I thought Apple might do something along the lines of:

A = 12 months​

M = 18 months​

M Pro/Max/Ultra = 24 months​

M Mega = 30 months​

With the huge caveat of +/- a standard deviation or two, for the M chips.

I saw the same data before, and IMO the most instructive thing, is how inconsistent it is. Next look at how slowly Apple updated various Macs.

I think Apple is in a relative rush right now to complete the transition, but long pauses in the Macs of the future seem very likely, just like there were in the past.

 

[moinmoin]

Next look at how slowly Apple updated various Macs.

With the huge caveat that those were dependent on Intel, which they no longer are. Who knows what plans Apple had with Macs that Intel's constant node troubles delayed and likely cancelled.

 

[Heartbreaker]

With the huge caveat that those were dependent on Intel, which they no longer are. Who knows what plans Apple had with Macs that Intel's constant node troubles delayed and likely cancelled.

There were still newer, slightly better chips. Apple was in no hurry.

 

[repoman27]

With the huge caveat of +/- a standard deviation or two, for the M chips.

I saw the same data before, and IMO the most instructive thing, is how inconsistent it is. Next look at how slowly Apple updated various Macs.

I think Apple is in a relative rush right now to complete the transition, but long pauses in the Macs of the future seem very likely, just like there were in the past.

At right around 40 million units for basic M series devices (M1, M2, etc.), annual cadence isn't crazy at all. It actually seems somewhat likely. The AX series was only in iPads and averaged less than 16 months between releases. The Pro/Max/Ultra/Whatever are probably a different story.

I presume Apple will continue to revise chassis designs on a 3 to 6 year cycle to amortize the tooling and associated costs. Once they settle into a product line-up that is entirely designed around Apple silicon, they can essentially refresh as often as they like. Apple is 100% in control of the M series production schedule, and those SoCs replace:

• platform and adjacencies (CPU, PCH, Thunderbolt controllers) from Intel
• discrete GPUs and associated DRAM subsystems from AMD or NVIDIA
• I/O controllers and associated DRAM subsystems (T1, T2, SSD controllers) either Apple internal or from Samsung, Toshiba, or SanDisk

Which is pretty much all of the stuff that matters in terms of a refresh and can be updated without changing the chassis design.