Discussion Apple Silicon SoC thread

Page 129 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Eug

Lifer
Mar 11, 2000
23,583
996
126
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:

Screen-Shot-2021-10-18-at-1.20.47-PM.jpg

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:


M1 Ultra discussion here:


M2 discussion here:


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:

 
Last edited:

fkoehler

Member
Feb 29, 2008
193
145
116
So basically Apple ran into ST performance wall for this architecture. M1 at least had excuse of X64 transition and extra work to enable it. With M2 stagnation is now obvious.

From what I've read, Apple has lost a LOT of engineering staff in the past several months. Their Back to the Office recently cost them their top dog Dir of Machine Learning.
 

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
Has it been confirmed these are Gent A15 cores (Avalanche-performance, Blizzard-efficiency) ?


I don't think we'll know until someone gets their hands on one. I just posted that assumption based on the performance gains and especially the fact it was claimed to be on 'second generation 5nm'. That means N5P, not N4 as some here claim as that is the 3rd generation of the N5 family! I have seen nothing conclusive that confirms or denies that belief.

The mystery is why they would use A15 cores instead of A16 which they obviously could have - the M2 tapeout deadline would be pretty much the same as A16's which ships only two months later. I will again point out the odd rumor that iPhone 14 Pro models will be the only ones getting A16 with the non-Pro getting A15.

It seems unlikely Apple would suddenly decide "hey we should market segment based on SoC generation", especially when they just released a $429 iPhone SE with the A15 SoC a few months ago. If you're gonna segment, you don't give the entry level model the same SoC as models that cost twice as much will get six months later!

So I am forced to draw the conclusion that A16 costs a fair bit more than A15 to make, or is limited in supply. The reason behind A16 costing more or having limited supply is probably also the reason M2 gets A15 cores instead of A16.
 
  • Like
Reactions: Tlh97 and blckgrffn

repoman27

Senior member
Dec 17, 2018
342
488
136
I don't think we'll know until someone gets their hands on one. I just posted that assumption based on the performance gains and especially the fact it was claimed to be on 'second generation 5nm'. That means N5P, not N4 as some here claim as that is the 3rd generation of the N5 family! I have seen nothing conclusive that confirms or denies that belief.

The mystery is why they would use A15 cores instead of A16 which they obviously could have - the M2 tapeout deadline would be pretty much the same as A16's which ships only two months later. I will again point out the odd rumor that iPhone 14 Pro models will be the only ones getting A16 with the non-Pro getting A15.

It seems unlikely Apple would suddenly decide "hey we should market segment based on SoC generation", especially when they just released a $429 iPhone SE with the A15 SoC a few months ago. If you're gonna segment, you don't give the entry level model the same SoC as models that cost twice as much will get six months later!

So I am forced to draw the conclusion that A16 costs a fair bit more than A15 to make, or is limited in supply. The reason behind A16 costing more or having limited supply is probably also the reason M2 gets A15 cores instead of A16.
I did do a quick visual comparison of the cores from Apple's M2 die images and the A15 die image from TechInsights (which I posted above), and they appear to be extremely similar.

It obviously makes sense that if the M1 uses the same cores and process as the A14, the M2 might use the same cores and process as the A15, especially if it were to be released at any point in the Oct 2021 to Aug 2022 timeframe. A July release might mean that it's arriving later than Apple originally planned. OTOH, they're also refreshing their two most affordable and best selling laptops right on time to nail the back-to-school window rather than waiting until October and missing the boat. This could be supply-chain related, or they may have had to delay the introduction of the M2 because the M1 Ultra was behind schedule due to manufacturing complexity. Who knows?
 

Eug

Lifer
Mar 11, 2000
23,583
996
126
It obviously makes sense that if the M1 uses the same cores and process as the A14, the M2 might use the same cores and process as the A15, especially if it were to be released at any point in the Oct 2021 to Aug 2022 timeframe. A July release might mean that it's arriving later than Apple originally planned. OTOH, they're also refreshing their two most affordable and best selling laptops right on time to nail the back-to-school window rather than waiting until October and missing the boat. This could be supply-chain related, or they may have had to delay the introduction of the M2 because the M1 Ultra was behind schedule due to manufacturing complexity. Who knows?
Some of the supply chain gurus are claiming the Mac Pro and Mac mini have been delayed. Previous predictions were for mid to late 2022, but now some are even saying 2023. This may partially just be educated guessing based on general supply chain trends as opposed to firm supply chain leaks, but regardless, supply chain info has not been very optimistic about these lower selling Macs.

If they're going to prioritize anything, it's going to be the laptops, and amongst the laptops, it's going to be the MacBook Air, since the MacBook Air is Apple's best selling Mac. They're going to sell a sheetload of these during the 2022 Back-To-School promotion, which should start next month, right about when the MacBook Air goes on sale.

Anyhow, given that the Mac Pro hasn't been updated since 2019, that means if it gets updated in late 2022 it would have been over 3 years, and if in early 2023 it would be close to 4 years. In that context, and because the SoC design is likely somewhat different, and because Apple has said M1 Ultra is the last of the M1 series, I think that provides more support for the idea that the Mac Pro won't be using M anything. Instead, as some of us have suggested previously, it may have a different nomenclature entirely. From a marketing perspective, it could be beneficial for Apple to have a different naming scheme for the SoCs used in a desktop that gets updated so infrequently.
 

Roland00Address

Platinum Member
Dec 17, 2008
2,196
260
126
I am away from a device with a keyboard to double check myself, but isn’t the single threaded geekbench like 10% generation over generations increase only via architecture once you take out the frequency improvements.. for something like A10 to A15? Like 1.1 to the 5th power due to 5 new generations is about 61% performance uplift?
 

mikegg

Golden Member
Jan 30, 2010
1,740
406
136
Oh, please don't reiterate silly internet nonsense. Some of the most reliable numbers available for A14 vs A15 performance differences are Anandtech's SPEC 2017 numbers.

For single-threaded, the Avalanche performance cores in the A15 demonstrated a 12.89% performance increase with a 8.07% higher maximum clock speed while using 13.87% less power than the Firestorm cores in the A14. The Blizzard efficiency cores in the A15 offered a 24.51% performance increase with a 10.59% higher maximum clock speed while using 3.33% more power than the Icestorm cores in the A14. Both chips were also using LPDDR4X-4266 with a 64-bit interface, and the process difference is only TSMC N5 to N5P.

If Intel or AMD achieved similar gains, would anyone consider it "too negligible to mention"?
Yep.

And by all reports, the M2 was ready back in the Spring Event already but the new Air design wasn't ready yet.

This means Apple is probably planning a yearly update cadence for the base M. This is actually a lot of improvements for each year - far more than what Intel and AMD could do.
 

Eug

Lifer
Mar 11, 2000
23,583
996
126
Yep.

And by all reports, the M2 was ready back in the Spring Event already but the new Air design wasn't ready yet.

This means Apple is probably planning a yearly update cadence for the base M. This is actually a lot of improvements for each year - far more than what Intel and AMD could do.
This doesn’t sound right. I’d be very surprised if M series is yearly.
 
  • Like
Reactions: scineram

Heartbreaker

Diamond Member
Apr 3, 2006
4,222
5,224
136
This doesn’t sound right. I’d be very surprised if M series is yearly.

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.
 
  • Like
Reactions: SpudLobby

Heartbreaker

Diamond Member
Apr 3, 2006
4,222
5,224
136
I am away from a device with a keyboard to double check myself, but isn’t the single threaded geekbench like 10% generation over generations increase only via architecture once you take out the frequency improvements.. for something like A10 to A15? Like 1.1 to the 5th power due to 5 new generations is about 61% performance uplift?

If you look at A-chip generational gains have been steadily trending downward. You can't keep making huge gains every generation.

A15 looks a lot like Zen4. It's essentially the same core with more cache and clock speed.

Nothing wrong with that, you just have to adjust expectations to reality.
 

Heartbreaker

Diamond Member
Apr 3, 2006
4,222
5,224
136
I think that provides more support for the idea that the Mac Pro won't be using M anything. Instead, as some of us have suggested previously, it may have a different nomenclature entirely. From a marketing perspective, it could be beneficial for Apple to have a different naming scheme for the SoCs used in a desktop that gets updated so infrequently.

I don't care much about naming, but economic reality dictates Mac Pro must be built from the same tape-out as the higher volume mid-high parts (Max-Ultra). While I don't care much about naming, but I expect they will use M naming if they are using an M tapeout. The use the M naming in the iPad even though presumable the M designation stands for Mac Chips. So for consistency. All M tapeouts will keep the M naming.

The need to use higher volume parts is why I keep thinking the eventual Mac Pro will be some kind of backplane design with M-Ultra blades.
 
  • Like
Reactions: Tlh97 and ashFTW

ashFTW

Senior member
Sep 21, 2020
302
225
96
The need to use higher volume parts is why I keep thinking the eventual Mac Pro will be some kind of backplane design with M-Ultra blades.
I have loved the idea of a computer built like that for a very long time. But I would also like "disaggregated" blades along with the M-Ultra blades. For example, a dedicated GPU or CPU or even a memory blade. This would require some fundamental rethink.
 

Heartbreaker

Diamond Member
Apr 3, 2006
4,222
5,224
136
For example, a dedicated GPU or CPU or even a memory blade. This would require some fundamental rethink.

And many new unique low volume parts, so that's not going to happen.

I'm just trying to think how they ramp up computing power, but with component reuse from lower end models.

The only thing I can come up with is Multiple Ultras.

4 Ultras with upgraded with double capacity LPDDR5x packages would have 256GB RAM per ultra, so 1TB of memory possible (though with significant NUMA caveats).

Downside is of course, you need software to recognize non uniform access to all the resources on other blades.

But a lot of the workloads that benefit from more than single Ultra performance levels are getting into Render Farm activities.

Even back before the Mac transition was announced there was ARM speculation, but I was always skeptical because of the Mac Pro issue. It simply doesn't make sense to do a unique tape-out for a low volume computer like the Mac Pro.

So do they just kill the Mac Pro, or do they somehow cobble together many lower end parts to build it... It now seems like the latter is likely but the topology is still a BIG unknown.

I've never been so curious, for so long, about a computer that I would never buy.

Hope it arrives later this year so we finally get to see how they are going to pull this off.
 

Eug

Lifer
Mar 11, 2000
23,583
996
126
I don't care much about naming, but economic reality dictates Mac Pro must be built from the same tape-out as the higher volume mid-high parts (Max-Ultra). While I don't care much about naming, but I expect they will use M naming if they are using an M tapeout. The use the M naming in the iPad even though presumable the M designation stands for Mac Chips. So for consistency. All M tapeouts will keep the M naming.
The iPad uses the same M designation because it's the exact same chip. If you think the Mac Pro would use an existing M series chip then yeah it makes sense that the Mac Pro would keep the M naming. However, there appears to be debate here whether or not the SoC in the Mac Pro would be different.

The need to use higher volume parts is why I keep thinking the eventual Mac Pro will be some kind of backplane design with M-Ultra blades.
I have loved the idea of a computer built like that for a very long time. But I would also like "disaggregated" blades along with the M-Ultra blades. For example, a dedicated GPU or CPU or even a memory blade. This would require some fundamental rethink.
Who would actually use this though? This doesn't seem to be the traditional Mac market. Or are you expecting Apple to go after a new market?

So do they just kill the Mac Pro, or do they somehow cobble together many lower end parts to build it... It now seems like the latter is likely but the topology is still a BIG unknown.
Apple has already said publicly the Mac Pro is coming.

I've never been so curious, for so long, about a computer that I would never buy.

Hope it arrives later this year so we finally get to see how they are going to pull this off.
Hahah. Same. The only reason I own 2 Mac Pros is because they are 1.5 decades old, and I got one for a 6-pack of beer, and another for $100 off Kijiji. :)
 

Heartbreaker

Diamond Member
Apr 3, 2006
4,222
5,224
136
The iPad uses the same M designation because it's the exact same chip. If you think it's the same chip then yeah that makes sense. But there appears to be debate here whether or not the SoC in the Mac Pro would be different.

I don't think you can make a business case for unique tape-out just for the Mac Pro. The volume is too low and modern tapeouts are too expensive.
 

ashFTW

Senior member
Sep 21, 2020
302
225
96
I don't think you can make a business case for unique tape-out just for the Mac Pro. The volume is too low and modern tapeouts are too expensive.
They can build servers like this, and use them internally for now. They could get into newer markets like cloud computing in the future. Sky is the limit. We are still in the infancy of computing.
 

trivik12

Senior member
Jan 26, 2006
252
205
116
They can build servers like this, and use them internally for now. They could get into newer markets like cloud computing in the future. Sky is the limit. We are still in the infancy of computing.
Apple has already moved away from their own DC. They use AWS heavily for internal purpose. I doubt they will go into capital intensive cloud computing requiring them to support all product/application stacks etc. If anything that will need a new spinoff company.
 

ashFTW

Senior member
Sep 21, 2020
302
225
96
And many new unique low volume parts, so that's not going to happen.

I'm just trying to think how they ramp up computing power, but with component reuse from lower end models.

The only thing I can come up with is Multiple Ultras.

4 Ultras with upgraded with double capacity LPDDR5x packages would have 256GB RAM per ultra, so 1TB of memory possible (though with significant NUMA caveats).
That's a great start. But doesn't mean things won't evolve into bigger better things over time. Right now, M1-Ultra is very GPU heavy. For my workloads, I would rather have a lot more CPU cores and minimal GPU. Others may want even more GPU. If a single architecture can satisfy the entire range of needs, that would be better.

Downside is of course, you need software to recognize non uniform access to all the resources on other blades.
Yeah, they abandoned NUMA a while ago. They probably have a ton of work to even make the OS scale beyond a certain number of cores.

But a lot of the workloads that benefit from more than single Ultra performance levels are getting into Render Farm activities.
And Ultra does make for a great node in this render farm. Future could easily progress like that, and there is less need for fatter nodes. I personally like to have as many cores I can get.

It now seems like the latter is likely but the topology is still a BIG unknown. I've never been so curious, for so long, about a computer that I would never buy.

Hope it arrives later this year so we finally get to see how they are going to pull this off.
Me too! And, I'm sure lots of hardware enthusiasts and professionals alike, whether they like  or not, are in the same boat.
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
N4 is still N5 process family—it's just a 6% optical shrink plus additional enhancements.

Apple is intimately aware of TSMC's progress in terms of process. They knew full well over two years ago that products based on N3 wouldn't be shipping in 2022, and the A16 was designed practically from the outset with N4 and a September 2022 release as the target. If Apple's silicon team knew all along that the A series chips wouldn't move to N3 before September 2023, do you really think they thought a bigger M series sized chip on N3 would happen before then?

Being the most influential customer of the foundry with the best available manufacturing process is not actually a problem for Apple's silicon ambitions.
TSMC'S original road map put N3 in 2022 (the public version). So yeah, Apple had to scramble a bit and make A15 an iteration of A14; rather than a more advanced A16 design. So it has affected the bulk of their product line. And of course Apple knew far in advance of almost anyone else given the fact that they have engineers at TMSC that work with FAB giant's Process development group.
 

Commodus

Diamond Member
Oct 9, 2004
9,206
6,799
136
This doesn’t sound right. I’d be very surprised if M series is yearly.

I wouldn't be. Apple sometimes takes longer for generational follow-ups. It wasn't until September 2016 that the Apple Watch leapt from the S1 to the S2, for example. Whether or not internal upheavals and shortages affected the timing, it's clear Apple is still building out its computer platform. It's entirely possible that Apple is just settling into a more regular cadence.

Remember, there's a credible rumor from Bloomberg's Mark Gurman that Apple will have an M3-based iMac next year... if so, you're likely looking at roughly 1-1.25 years between updates to the underlying chip design.
 
  • Like
Reactions: mikegg

ashFTW

Senior member
Sep 21, 2020
302
225
96
Apple has already moved away from their own DC. They use AWS heavily for internal purpose. I doubt they will go into capital intensive cloud computing requiring them to support all product/application stacks etc. If anything that will need a new spinoff company.
You are right about them using AWS heavily for their current services, most of which do not require heavy graphics and ML. As Apple moves into new markets like AR/VR, and autonomous cars, they will have a tremendous increase in their need for backend Graphics and ML. Apple has been populating the client hardware with incredible capability, but they know what special hardware they need on the server side to provide amazing experiences, and they are building towards that goal. It will then become a core competency, and that's not something you outsource or spinoff.
 

Heartbreaker

Diamond Member
Apr 3, 2006
4,222
5,224
136
Remember, there's a credible rumor from Bloomberg's Mark Gurman that Apple will have an M3-based iMac next year... if so, you're likely looking at roughly 1-1.25 years between updates to the underlying chip design.

But that might just be skipping M2 in the midrange and above. So could be about 2 years since the last upper range chip.

The point here is cadence of tapeouts for the product ranges.

You can do an entry chip every ~18month average (or even shorter) since it's also the iPad chip. This is the highest volume M chips, so that's no problem at all.

But once you hit midrange, volume drops a lot, so cadence likely will as well. I'd expect a 2 year cadence average or longer.
 

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
But that might just be skipping M2 in the midrange and above. So could be about 2 years since the last upper range chip.

The point here is cadence of tapeouts for the product ranges.

You can do an entry chip every ~18month average (or even shorter) since it's also the iPad chip. This is the highest volume M chips, so that's no problem at all.

But once you hit midrange, volume drops a lot, so cadence likely will as well. I'd expect a 2 year cadence average or longer.


They can't skip M2 for the high end, that's what the Mac Pro will be based on. The M1 Max die is able to link to only one other M1 Max to create the M1 Ultra. They need an M2 Max with three off chip I/O pads to create the "M2 Extreme" that will go in the Mac Pro. They can't build it out of M1.

Given that M1 had only two mask sets - the M1 and the M1 Pro/Max which was the same die just "chopped" when making M1 Pros I don't see any reason why they can't have the full line for each generation. If they were doing Mx yearly I could see them only doing the the Pro/Max every other year, but this has been more like 18 months as the rumors claimed. If they maintain that there's no way they go three years between new Pro/Max/Ultra/Extreme offerings.
 

Mopetar

Diamond Member
Jan 31, 2011
7,797
5,899
136
For some something like a MacBook Air a yearly fresh makes sense, since it's a consumer product. Even if the performance gains are meager, the changes to the hardware can help carry the product.

Professional machines and their users are less concerned with small performance bumps or hardware changes and people won't have a problem buying year old hardware as long as they know it's going to receive long-term support.

I'm more interested in where Apple goes from here than what we see in terms of high-end M2 offerings. We can already make an educated guess of what an M2 Max or Ultra could be like. I want to see how they approach the problem of scaling designs across their entire product line.
 

StinkyPinky

Diamond Member
Jul 6, 2002
6,761
777
126
Honestly, I'm a fan of the M series chips and I have a M1 Max laptop myself.

However my wife hates macs so I bought her a Samsung Book 2 or whatever it is called. Fantastic laptop. Came with 16GB ram, i7 10 core cpu, 512GB SSD, Full HD OLED display on 13 inches, great battery life that lasts forever, with good port choice and well made, and it supports four external displays. Cost me a grand. Apple need to stay competitive.

I have to be critical of Apple here. There is no way a modern laptop that costs that much money should support one display. That's just not acceptable.
 
Last edited:
  • Like
Reactions: gdansk