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]

More SKUs, more qualification, more complexity for repair/warranty. There's a reason why companies selling longer lived products will stick with the same design even when cheaper/better parts become available, unless the cost difference becomes large enough. And when they do it either gets sold under a different name (i.e. like consoles where it is still a "PS5" but they change the form factor and everyone knows if it is the original version or a newer version) or it'll be done as a rev (i.e. all the wireless routers that have rev 1, rev 2, rev 3 over the years that they sell the "same" router and while you might think you can treat them the same you'll see stuff like "only rev 1 and 2 can install DD-WRT, rev 3 uses a different SoC that it isn't compatible with")

I get where you’re coming from. However, correct me if I’m wrong, but I believe the parts were largely interchangeable. For example, the new 32 nm model A5 iPad had the same battery size as the 45 nm model, and so it got longer battery life, and it shared other components too.

There was no way to tell which version you got unless you turned it on and installed a third party app.

So, if the housing is the same, the screen is the same, the battery is the same, and the speakers are the same, what else is there? The logic board with the SoC is just one piece. You can’t swap parts on the logic board. In fact, that’s not how Apple approached repairs anyway. Apple’s “fix” for a broken iPad is to give you a new iPad and recycle the old one.

And iPads sell a lot less volume than iPhones.

 

[eek2121]

IDK. I think it might be worthwhile if they're segmenting Pro/NonPro for the leading/last year A-series chips and anticipate A17 sales going well into 2024, 2025 (might use it in an iPad too). Might also be in time for the next SE. Remember, they still sell the iPhone 13 & the 12, and the SE uses the A15.

N3E is certainly the better node though (overall, put the SRAM aside vs cost, power, yields) what with FinFlex, the superior yields and modest performance boost. I don't think it'll be a big deal, though it'll certainly have improved power characteristics vs the N3B A17.

I'm skeptical though b/c of the tapeout costs, guess we'll see.

Remember that Apple will make these chips for many years. They will likely still be making the M1 2-3 years from now, for example. The M1 is almost 3 years old.

I don’t claim that the rumor is correct, but they will likely easily recoup the costs.

 

[SpudLobby]

Remember that Apple will make these chips for many years. They will likely still be making the M1 2-3 years from now, for example. The M1 is almost 3 years old.

I don’t claim that the rumor is correct, but they will likely easily recoup the costs.

You are restating what I am saying, haha. (I argue that this is in fact a reason they might want to do a new tapeout, which is why I mentioned the SE & the A15, or that Apple still sell the iPhone 12 & 13 with their A14 and A15 SoCs.) I mention the tapeout costs because that calculus — does it pay off given superior cost structures, yield in N3E — and given Apple’s depreciation strategies — is what is explicitly in question. We basically just don’t know for sure.

It makes sense to me though, I don’t think the rumor seems off entirely. The fact that they moved to vanilla N4 over N5P for the very paltry update that was the A16 suggests Apple do benefit from using the most cost optimized process where they can, and N4 isn’t fully design rule compatible with N5 to my understanding.

 

[SpudLobby]

If they do another SE they would have strong incentive to use A16, since while N3E will be cheaper than N3B it will still be more expensive than N4. Heck, they might even do a special version of A16 that uses LPDDR4X if there's still a meaningful difference in its price versus LPDDR5 since there's a lot of pressure on the BOM to make the SE's price point while preserving margins.

I think you and latched on to one part of my post. It doesn’t have to be an SE alone. Forget that thing.

Remember, Apple updates the Pros with the latest chip now, and base models keep last year’s chip. I doubt this will change next year, it’s here to stay by most reports. So the iPhone 16 base models in 2024 will use an A17.

Again, they still sell the iPhone 12 and 13, too, so that’s another pattern (A14’s and A15’s are still in production and I suspect they will keep a similar pattern for old base iPhones in 2023 and 2024, those then being the 13, 14, and 14, 15) and they now diverged the chip lineup between base flagship and Pro phones. You don’t need the SE to make this work. Come on guys, there are multiple inflection points.

 

[Eug]

You are restating what I am saying, haha. (I argue that this is in fact a reason they might want to do a new tapeout, which is why I mentioned the SE & the A15, or that Apple still sell the iPhone 12 & 13 with their A14 and A15 SoCs.) I mention the tapeout costs because that calculus — does it pay off given superior cost structures, yield in N3E — and given Apple’s depreciation strategies — is what is explicitly in question. We basically just don’t know for sure.

It makes sense to me though, I don’t think the rumor seems off entirely. The fact that they moved to vanilla N4 over N5P for the very paltry update that was the A16 suggests Apple do benefit from using the most cost optimized process where they can, and N4 isn’t fully design rule compatible with N5 to my understanding.

Yeah, Apple sells almost a quarter billion iPhones a year. Not quarter billion dollars, but a quarter billion iPhones, at almost $1000 a pop, or over $200 billion a year. That's a lot chips they can amortize over.

And that's not even counting iPads, etc. If you count all the A-series devices, that's getting closer to 300 million SoCs per year. The sold over 60 million iPads last year, although that includes some M series models.

 

[Glo.]

Yeah, Apple sells almost a quarter billion iPhones a year. Not quarter billion dollars, but a quarter billion iPhones, at almost $1000 a pop, or over $200 billion a year. That's a lot chips they can amortize over.

And that's not even counting iPads, etc. If you count all the A-series devices, that's getting closer to 300 million SoCs per year. The sold over 60 million iPads last year, although that includes some M series models.

Almost 250 mln iPhones, over 100 mln iPads = 350 mln A and M chips.

And on top of that - you have Mac, AppleWatch, which also has an SOC, and all of the other smaller SOCs that Apple designed. We are talking about a number of Apple SOCs that goes into almost half a billion, per year, chips.

 

[Doug S]

It makes sense to me though, I don’t think the rumor seems off entirely. The fact that they moved to vanilla N4 over N5P for the very paltry update that was the A16 suggests Apple do benefit from using the most cost optimized process where they can, and N4 isn’t fully design rule compatible with N5 to my understanding.

But A16 was a totally new design. It might not have made much of an impression given how similar the A14/A15/A16 cores were (except for under the hood stuff like number of physical address bits) and the major jump for the GPU they were planning had to be backed off and last year's GPU used so yeah it was underwhelming. Still, it wasn't the exact same SoC getting ported from N5P to N4 for the heck of it, and it did have some new stuff like LPDDR5 controllers. They targeted N4 because that was the best available process for the timeline they were shipping A16.

They targeted N3B (though I still think what Apple's chips are made with may be have some elements of N3E because it has been a half year since TSMC went into "mass production" and TSMC has done node tweaks for AMD so they certainly would for Apple!) because it was the best available process at the time A17 was shipping.

And AFAIK I thought N4 was design rule compatible with N5, but that's irrelevant to my point since A16 was a brand new SoC so they weren't porting an existing design.

 

[SpudLobby]

But A16 was a totally new design. It might not have made much of an impression given how similar the A14/A15/A16 cores were (except for under the hood stuff like number of physical address bits) and the major jump for the GPU they were planning had to be backed off and last year's GPU used so yeah it was underwhelming. Still, it wasn't the exact same SoC getting ported from N5P to N4 for the heck of it, and it did have some new stuff like LPDDR5 controllers. They targeted N4 because that was the best available process for the timeline they were shipping A16.

They targeted N3B (though I still think what Apple's chips are made with may be have some elements of N3E because it has been a half year since TSMC went into "mass production" and TSMC has done node tweaks for AMD so they certainly would for Apple!) because it was the best available process at the time A17 was shipping.

And AFAIK I thought N4 was design rule compatible with N5, but that's irrelevant to my point since A16 was a brand new SoC so they weren't porting an existing design.

Hmm so it looks like there are some minor design rule changes, but not anything major. N5 and N4 being design rule compatible is what I had originally thought - much like N7 and N6 (albeit the shrink there with EUV was much larger vs the 6% optical shrink from N5 to N4 and BEOL enhancements).

My bad on this, you are correct. I suppose I thought they did make some nnonzero changes - from WikiChip "N4. N4 is a derivative of N5 (which is TSMC includes it under the “5nm family”) very similar to the N6 node they introduced to enhance their N7 node. N4 is said to provide a small 6% die shrink through “standard cell innovation” and design rule changes that help realize better area efficiency.". Maybe that's merely referring to the optical shrink itself in paraphrased terms.

Moving on:

and the major jump for the GPU they were planning had to be backed off and last year's GPU used so yeah it was underwhelming. Still, it wasn't the exact same SoC getting ported from N5P to N4 for the heck of it, and it did have some new stuff like LPDDR5 controllers.

Yes, I know all of this. I mean the core CPU, NPU IP was fairly similar, though - primarily about new physical layouts. Again matter of degree.

At any rate, this is still peripheral to a very simple question: Do they have the sustained scale and market to possibly reap benefits? As I keep pointing out, they do produce their old SoC's for some time forward. If N3B really is lacking in some way or another RE: yields and they think a new tapeout is justified, it's not the craziest thing they've ever done. We just do not know what the margins for them look like wrt the fixed costs, but again:

If there is a significant cost decrement, amortizing the tapeout wouldn't be crazy at all, because they've shifted towards keeping older chips in their lineup to the extent that brand new flagship iPhones now arrive with last year's chip. Even if the A16 had gone as planned, there is no way that was coming to the iPhone 14 (non-Pro). The trendline is clear.

Doug: I totally agree it makes far, far more sense with a new design and the N3B to N3E jump is not like N5 to N4. I will be blunt, I think the rumor isn't especially likely. Though I also don't think it's totally crazy given their economies of scale and what we've heard about N3B and TSMC with N3, N3E is very explicitly marketed as a cost-optimized version of N3, which afaict is basically done for and N3E + N3S are the way forward. Plus, they don't have crazy aversions to a few tweaks to one SoC's name - even for binning - the iPad Mini's A15 is capped to 3GHz for example. We also see the A15 4 and 5-core units with the iPhones in 2021, and now we see full segmentation in 2022 will likely continue, which means an A17 will be around for some time.

 

[SpudLobby]

Yeah, Apple sells almost a quarter billion iPhones a year. Not quarter billion dollars, but a quarter billion iPhones, at almost $1000 a pop, or over $200 billion a year. That's a lot chips they can amortize over.

And that's not even counting iPads, etc. If you count all the A-series devices, that's getting closer to 300 million SoCs per year. The sold over 60 million iPads last year, although that includes some M series models.

Exactly. I'll say I've backed off slightly from this but I mean, given what we know about Apple's chip strategy and economies of scale, TSMC and N3 specifically and N3E, it's not impossible

 

[Doug S]

The question I would ask is, if Apple would do this to capture the cost savings why wouldn't they have for example starting making A15s in N4? Between non Pro iPhone 14s and iPhone 13s still being sold that's probably going to be easily 100+ million A15s installed in iPhones sold between last September and whenever the last device containing A15 is sold by Apple. Surely that's enough to amortize new masks, and definitely is enough to just do the optical shrink and increase your chips per wafer even if the per wafer cost isn't less (but from everything I've heard, it IS less for N6 vs N7/N7P and N4 vs N5/N5P)

They might save more money with the N3E transition, but they don't seem to have been interested in saving money like this in the past.

 

[SpudLobby]

The question I would ask is, if Apple would do this to capture the cost savings why wouldn't they have for example starting making A15s in N4? Between non Pro iPhone 14s and iPhone 13s still being sold that's probably going to be easily 100+ million A15s installed in iPhones sold between last September and whenever the last device containing A15 is sold by Apple. Surely that's enough to amortize new masks, and definitely is enough to just do the optical shrink and increase your chips per wafer even if the per wafer cost isn't less (but from everything I've heard, it IS less for N6 vs N7/N7P and N4 vs N5/N5P)

They might save more money with the N3E transition, but they don't seem to have been interested in saving money like this in the past.

Fair.

 

[soresu]

They might save more money with the N3E transition, but they don't seem to have been interested in saving money like this in the past.

Node costs have increased, mask costs have increased, and now Apple's own chip designs per generation have increased for Mx and likely some future dedicated Vision headset SoC too.

At a certain point something has to give - especially if growth isn't quite what it used to be in some segments.

 

[Doug S]

Node costs have increased, mask costs have increased, and now Apple's own chip designs per generation have increased for Mx and likely some future dedicated Vision headset SoC too.

The increase in per generation cost is irrelevant to the extent those additional costs are supporting new products, which are priced to easily pay for it. Apple has scale like no one else, increases in mask costs aren't concerning for them (but if they were the solution wouldn't be needing more mask sets) as that cost is a fraction of a percent of the sales price for an iPhone/Mac/Vision Pro.

Apple is benefiting from re-using the same CPU/GPU/NPU cores and just varying their number and clock rate to meet power/performance needs. So yeah they need to do a new floorplan for the M3 SoC vs A17 SoC and the Vision Pro SoC will be yet another floorplan, but 90% of the engineering work has already done in designing the four main cores (2x CPU, GPU, NPU)

This option has been available for a while now to save money, but they have never done it. As you say node/wafer costs are increasing so maybe it has finally reached the point where Apple feels it is worth doing. If they do it will have to do with those wafer prices increasing and offering bigger potential savings from a shrink/port than having anything to do with Apple having additional design cost for the Mac and Vision lines.

 

[ashFTW]

Maybe N3B is a short lived node and everything is required to move to N3E etc and the resources currently allocated to the N3B production will get allocated elsewhere. Maybe that’s why Apple is forced to switch A17 in next year. What about M3* chips — will they be on N3B or N3E? If N3E, will that introduce some delay so we may not see any in ’23?

 

[Doug S]

Maybe N3B is a short lived node and everything is required to move to N3E etc and the resources currently allocated to the N3B production will get allocated elsewhere. Maybe that’s why Apple is forced to switch A17 in next year. What about M3* chips — will they be on N3B or N3E? If N3E, will that introduce some delay so we may not see any in ’23?

Hmmm if they want to kill N3B as quickly as possible that might explain why they were willing to make concessions to Apple like paying for KGD instead of per wafer. If that's the case they would almost certainly be paying Apple's transition/porting costs from N3B to N3E as well.

 

[eek2121]

Yeah, Apple sells almost a quarter billion iPhones a year. Not quarter billion dollars, but a quarter billion iPhones, at almost $1000 a pop, or over $200 billion a year. That's a lot chips they can amortize over.

And that's not even counting iPads, etc. If you count all the A-series devices, that's getting closer to 300 million SoCs per year. The sold over 60 million iPads last year, although that includes some M series models.

Slight nitpick here, but iPhone prices start at $429. Only the Pro Max, or certain models with a ton of added storage space, get above $1,000.

 

[ashFTW]

Slight nitpick here, but iPhone prices start at $429. Only the Pro Max, or certain models with a ton of added storage space, get above $1,000.

iPhone ASP (Average Selling Price) has been in the 800-950 range for the last few quarters. And it seems to continue to grow ..

 

[eek2121]

iPhone ASP (Average Selling Price) has been in the 800-950 range for the last few quarters. And it seems to continue to grow ..

Note that ASP can be heavily weighted by folks like me who pick the most expensive stuff.

ASP != median selling price. ASP could be over $1,000, but that does not mean most users are paying even close to that much. Example: If 3 users buy a phone for $429 and 1 user buys the top $1,599 variant, ASP ends up being ~$722

I say this because the bulk of iPhone sales are well under $1,000.

 

[Eug]

Slight nitpick here, but iPhone prices start at $429. Only the Pro Max, or certain models with a ton of added storage space, get above $1,000.

Well, I said almost $1000 per iPhone. It's $205 billion for 225.3 billion iPhones.

That works out to ~$910 per iPhone.

iPhone ASP (Average Selling Price) has been in the 800-950 range for the last few quarters. And it seems to continue to grow ..

In other news, Mark Gurman is at it again, now claiming (again?) there is a 30"+ iMac in the pipeline.

Personally, my holy grail would be a 29.4" 5K iMac and standalone screen at 200 ppi, but that won't happen.

 

[ashFTW]

Note that ASP can be heavily weighted by folks like me who pick the most expensive stuff.

ASP != median selling price. ASP could be over $1,000, but that does not mean most users are paying even close to that much. Example: If 3 users buy a phone for $429 and 1 user buys the top $1,599 variant, ASP ends up being ~$722

I say this because the bulk of iPhone sales are well under $1,000.

Thanks for the math lesson that average is not usually equal to median /s

 

[Doug S]

One interesting thing I read today pointed out that Vision Pro will use M2, when being released early next year it could have used the M3 at least from a process perspective. But if M3 will always be N3E from day one then Vision Pro could not have used it because there are already working existing now and have probably existed for a while.

And if M3 is N3E, then a mid year transition of A17 to N3E makes perfect sense, because THAT would be the "real" A17 - the N3B version would be the one that's a port that will be discontinued as soon as N3E wafers start rolling off the line.

 

[Doug S]

Note that ASP can be heavily weighted by folks like me who pick the most expensive stuff.

ASP != median selling price. ASP could be over $1,000, but that does not mean most users are paying even close to that much. Example: If 3 users buy a phone for $429 and 1 user buys the top $1,599 variant, ASP ends up being ~$722

I say this because the bulk of iPhone sales are well under $1,000.

Apple reportedly sells more Pro than non Pro of the "new" iPhone model (i.e. iPhone 14 right now) so of the ones buying the latest and greatest, they are all paying over $1000. The people buying lesser are buying a mix of non Pro, last year's model, and the SE but given where the ASP lies I doubt the median is all that different.

 

[eek2121]

Apple reportedly sells more Pro than non Pro of the "new" iPhone model (i.e. iPhone 14 right now) so of the ones buying the latest and greatest, they are all paying over $1000. The people buying lesser are buying a mix of non Pro, last year's model, and the SE but given where the ASP lies I doubt the median is all that different.

Would not surprise me. The 256gb Pro Max has been my goto every year. Though thanks to the upgrade program I pay $699 per phone.

 

[soresu]

Would not surprise me. The 256gb Pro Max has been my goto every year. Though thanks to the upgrade program I pay $699 per phone.

Ooof, that's an expensive habit even so 🤑

It's sad that phone manufacturers are so damn cheap though - and that's not an Apple specific comment, it's an industry wide grift.

You can get a 2 TB external SSD for less than £100 these days - to charge so much for a mere 1/8th the space is truly miserly.

 

[Doug S]

Ooof, that's an expensive habit even so 🤑

It's sad that phone manufacturers are so damn cheap though - and that's not an Apple specific comment, it's an industry wide grift.

You can get a 2 TB external SSD for less than £100 these days - to charge so much for a mere 1/8th the space is truly miserly.

I'm not sure those NAND chips are necessarily comparable. Unlike in a phone, in a SSD power isn't too much of a concern.