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

 

[Roland00Address]

Also looks like the kind of small incremental upgrade, that Apple used to get from Intel.

Incremental upgrades are useful if you can deliver new ones every 12 to 18 months.

Incremental upgrades every 36 months, and thus two of them every 72 months is a different matter.

(intel was between those two I just listed, but one can argue it was closer to the second recently.)

 

[Doug S]

Also looks like the kind of small incremental upgrade, that Apple used to get from Intel.

I'll bet going from 64 to 96 GB has to do with the M2 generation using LPDDR chips that are 50% denser. The market for a Macbook Pro (I'm guessing that's what this is) with even larger memory above what they offer is not zero but small enough it probably isn't worth the hassle of offering larger SKUs - they'd have to manage that all the way through the manufacturing process. Likely extending to their DRAM supplier who I'm guessing is responsible for manufacturing the packages they are installing on the AS MCMs. That's the price paid for offering that insane memory bandwidth.

The only place where that might make sense is the Mac Pro, but even there we aren't yet sure if they will offer larger LPDDR packages to order models with more RAM. If not 384 GB will be it. Apple will have information about how much DRAM people are installing in Mac Pros currently (even if it is third party added post-purchase, for those who have their Macs sending diagnostic info to Apple) so they know the size of the addressable market.

 

[ashFTW]

I'll bet going from 64 to 96 GB has to do with the M2 generation using LPDDR chips that are 50% denser.

Don’t we already know this from M2 MacBook Air going to 24GB max from 16GB max in M1 Air?

 

[FlameTail]

Yes it does. It seems all great engineers went to Qualcomm via Nuvia.

And Rivos.

In all seriousness, I am sure they will be super interesting comparisons between Nuvia/Oryon designs and Apple's designs.

 

[Doug S]

Don’t we already know this from M2 MacBook Air going to 24GB max from 16GB max in M1 Air?

Yes they are likely using the same LPDDR packages across each Apple Silicon generation. AFAIK no DRAM OEM has 128 bit wide LPDDR packages on their standard parts list, so it probably helps Apple's cost and availability guarantees to have a single standard package used across the line.

 

[mikegg]

Also looks like the kind of small incremental upgrade, that Apple used to get from Intel.

Huh?

We've known for a long time what the M2 would do because the iPhone 13 came out 15 months ago. Anandtech gave the A15 a huge thumbs up in improvements.

M2 is a respectable upgrade, especially MT and GPU. The problem isn't the performance. The problem is that it's been delayed.

From reports and rumors, Apple seems to want to do a yearly update cadence but things have been delayed a lot due to supply chain, COVID work from home, and launching brand new Mac designs. If Apple can crank out these improvements yearly starting with the M3, then these are really good improvements.

M3 is the generation I expect a huge improvement from the M1 because of 3nm. The A16 already has 1874 in ST performance. At minimum, we can expect M3 to have 2,100 ST score.

If Apple is daring, they would skip the A16 and go straight to the A17 for M3 and use 3nm and launch it by Summer 2023. That'd allow Apple to take the ST lead again. This might be plausible because porting the A16 design to 3nm would be a big resource sink. It'd be smarter to just use the A17, which will be designed for 3nm from the start.

 

[Eug]

TSMC's Arizona Plant Will Manufacture 4nm Chips Starting in 2024 at Apple's Request

TSMC's $12 billion chip plant currently under construction in Arizona will begin producing 4-nanometer chips as soon as it opens in 2024...

www.macrumors.com

Apparently Mark Gurman says TSMC Arizona is starting with N4 at launch, and that Apple will have 1/3rd of TSMC’s eventual overall volume of 20000+ wafers per month.

BTW, assuming that M2 Max Geekbench leak is legit, what process do you think that is on? N5P? N4? As in not N3.

 

[mikegg]

TSMC's Arizona Plant Will Manufacture 4nm Chips Starting in 2024 at Apple's Request

TSMC's $12 billion chip plant currently under construction in Arizona will begin producing 4-nanometer chips as soon as it opens in 2024...

www.macrumors.com

Apparently Mark Gurman says TSMC Arizona is starting with N4 at launch, and that Apple will have 1/3rd of TSMC’s eventual overall volume of 20000+ wafers per month.

BTW, assuming that M2 Max Geekbench leak is legit, what process do you think that is on? N5P? N4? As in not N3.

It will be N5P because the M2 is on N5P I believe.

N4 will be useful to Apple for a long time because Apple will continue to make the A16 and use it in future iPads, Apple TVs, etc.

 

[Tigerick]

TSMC's Arizona Plant Will Manufacture 4nm Chips Starting in 2024 at Apple's Request

TSMC's $12 billion chip plant currently under construction in Arizona will begin producing 4-nanometer chips as soon as it opens in 2024...

www.macrumors.com

Apparently Mark Gurman says TSMC Arizona is starting with N4 at launch, and that Apple will have 1/3rd of TSMC’s eventual overall volume of 20000+ wafers per month.

BTW, assuming that M2 Max Geekbench leak is legit, what process do you think that is on? N5P? N4? As in not N3.

Arizona Fab will only start operating in 2024, so it is not related on M2 Pro/Max which I strong believe will be made by TSMC N3 process.

 

[FlameTail]

Huh?

We've known for a long time what the M2 would do because the iPhone 13 came out 15 months ago. Anandtech gave the A15 a huge thumbs up in improvements.

M2 is a respectable upgrade, especially MT and GPU. The problem isn't the performance. The problem is that it's been delayed.

From reports and rumors, Apple seems to want to do a yearly update cadence but things have been delayed a lot due to supply chain, COVID work from home, and launching brand new Mac designs. If Apple can crank out these improvements yearly starting with the M3, then these are really good improvements.

M3 is the generation I expect a huge improvement from the M1 because of 3nm. The A16 already has 1874 in ST performance. At minimum, we can expect M3 to have 2,100 ST score.

If Apple is daring, they would skip the A16 and go straight to the A17 for M3 and use 3nm and launch it by Summer 2023. That'd allow Apple to take the ST lead again. This might be plausible because porting the A16 design to 3nm would be a big resource sink. It'd be smarter to just use the A17, which will be designed for 3nm from the start.

Well, Apple's plan was to update the Mac processors once every 18 months.

So
M1 = A14 architecture
M2 = A15 architecture
M3 = A17 architecture.

Unless Apple has announced otherwise?

 

[Doug S]

Well, Apple's plan was to update the Mac processors once every 18 months.

So
M1 = A14 architecture
M2 = A15 architecture
M3 = A17 architecture.

Unless Apple has announced otherwise?

Apple never "announced" an 18 month cadence, that was reported as a rumor that everyone decided to believe. There's no plausible reason to have an 18 month cadence when the iPhone SoC that shares the same cores has a 12 month cadence. I'm guessing someone got hold of the planned M2 ship date, saw it was well over a year after M1 shipped and decided to infer an 18 month timeline based on two data points. Even if M3 comes out this time next year adding a third data point I would remain highly skeptical until early 2025 rolled around with no imminent announcement of M4.

I think schedule factors just worked out this way, between Apple having a lot of extra work launching the first ARM version of the various Mac product lines, development/testing timelines for the 2C and 4C variants, and TSMC altering their previous Q2 release cadence with N3. There's only such much engineering bandwidth, but since the subsequent versions will require less work they can tighten up the schedule later. Then add the pandemic, work from home, and ongoing supply chain delays on top of everything else. Heck the current lockdowns in China won't just affect iPhone production, they may end up delaying planned ship dates for M2 Pro/Max systems and M3 Macs also.

For instance, I think there's no way the original plan was to make M2 on N5 and M2 Pro/Max on N3 as rumored, but if that's how it ends up that's just a result of having to change plans to accommodate changing circumstances.

Now I could believe a 12 month cadence, a 24 month cadence, or even a 12 month cadence that sometimes skips the base or non base variants in a given generation. I just don't buy an 18 month cadence, because it makes no kind of sense at all.

 

[FlameTail]

Apple never "announced" an 18 month cadence, that was reported as a rumor that everyone decided to believe. There's no plausible reason to have an 18 month cadence when the iPhone SoC that shares the same cores has a 12 month cadence. I'm guessing someone got hold of the planned M2 ship date, saw it was well over a year after M1 shipped and decided to infer an 18 month timeline based on two data points. Even if M3 comes out this time next year adding a third data point I would remain highly skeptical until early 2025 rolled around with no imminent announcement of M4.

I think schedule factors just worked out this way, between Apple having a lot of extra work launching the first ARM version of the various Mac product lines, development/testing timelines for the 2C and 4C variants, and TSMC altering their previous Q2 release cadence with N3. There's only such much engineering bandwidth, but since the subsequent versions will require less work they can tighten up the schedule later. Then add the pandemic, work from home, and ongoing supply chain delays on top of everything else. Heck the current lockdowns in China won't just affect iPhone production, they may end up delaying planned ship dates for M2 Pro/Max systems and M3 Macs also.

For instance, I think there's no way the original plan was to make M2 on N5 and M2 Pro/Max on N3 as rumored, but if that's how it ends up that's just a result of having to change plans to accommodate changing circumstances.

Now I could believe a 12 month cadence, a 24 month cadence, or even a 12 month cadence that sometimes skips the base or non base variants in a given generation. I just don't buy an 18 month cadence, because it makes no kind of sense at all.

1. I never explicitly said that Apple announced an 18 month cadence, in my original reply.

Yeah, I daresay an 18 month cadence is indeed questionable.

Now, there is a question of whether it makes sense to update the M chips every year. Apple certainly has the resources to do so, but is it worth it?

 

[Mopetar]

Honestly, Apple could go 2 years between updates. If you just focus on making products that people will keep for 6+ years and support them, you don't need to worry about a yearly cadence. Anyone who's held on that long will either happily wait another year or buy a year old product without any remorse.

Even phones in general are reaching the point where upgrading every two years isn't necessary. There aren't really any components that are going to get so much better in that span of time where you just have to have it.

 

[Eug]

What's wrong with ~18 months? That was roughly the time frame for the iPad Pros, which used pre-M type chips.

The average between releases is over 500 days, and the very shortest is over 400 days.

2015-11 - A9X
2017-06 - A10X
2018-10 - A12X
2020-03 - A12Z
2021-04 - M1
2022-10 - M2

 

[Heartbreaker]

Honestly, Apple could go 2 years between updates. If you just focus on making products that people will keep for 6+ years and support them, you don't need to worry about a yearly cadence. Anyone who's held on that long will either happily wait another year or buy a year old product without any remorse.

Even phones in general are reaching the point where upgrading every two years isn't necessary. There aren't really any components that are going to get so much better in that span of time where you just have to have it.

Yearly never made any sense. They do that for iPhone, but they sell 10x has many iphones as Macs.

2 year cadence on average, makes sense, though I expect there will be no fixed cadence. They will do updates when it makes sense for Apple to do updates.

Even if they update chips, that doesn't mean the will update low priority products. Look at Mac Mini. That's already gone 2 over years without an update. Can't blame that on Intel...

 

[Doug S]

Yearly never made any sense. They do that for iPhone, but they sell 10x has many iphones as Macs.

2 year cadence on average, makes sense, though I expect there will be no fixed cadence. They will do updates when it makes sense for Apple to do updates.

Even if they update chips, that doesn't mean the will update low priority products. Look at Mac Mini. That's already gone 2 over years without an update. Can't blame that on Intel...

The main benefit of M2 over M1 is reduced power consumption, which is kind of pointless for the Mini.

 

[FlameTail]

The main benefit of M2 over M1 is reduced power consumption, which is kind of pointless for the Mini.

The main benefit of M2 over M1 is the huge jump in GPU performance.

 

[mikegg]

What's wrong with ~18 months? That was roughly the time frame for the iPad Pros, which used pre-M type chips.

The average between releases is over 500 days, and the very shortest is over 400 days.

2015-11 - A9X
2017-06 - A10X
2018-10 - A12X
2020-03 - A12Z
2021-04 - M1
2022-10 - M2

Those iPad chips were only used inside iPads. The base M chip will be used in many more devices: iPad Pro 11", iPad Pro 12.9", Macbook Air, iMac, 13" Macbook Pro, and the Mac Mini. Rumors are that Apple will release a M3 15" Macbook Air in 2023 to replace the 13" Macbook Pro. The base M chip could also be used in future devices such as VR Headset or a gaming-focused Apple TV.

The base M chip is far more important and much higher volume than previous iPad SoCs. The economy of scale suggests Apple will update the base M chip at a faster rate than the old X series.

 

[Eug]

Those iPad chips were only used inside iPads. The base M chip will be used in many more devices: iPad Pro 11", iPad Pro 12.9", Macbook Air, iMac, 13" Macbook Pro, and the Mac Mini. Rumors are that Apple will release a M3 15" Macbook Air in 2023 to replace the 13" Macbook Pro. The base M chip could also be used in future devices such as VR Headset or a gaming-focused Apple TV.

The base M chip is far more important and much higher volume than previous iPad SoCs. The economy of scale suggests Apple will update the base M chip at a faster rate than the old X series.

Apple took 20 months to update M1 to M2, which is actually slower than the old X series average.

Furthermore, if M3 arrives in June 2023, I’ll be quite surprised. I’m guessing somewhere between fall 2023 and spring 2024.

 

[eek2121]

I wouldn’t expect anything except modest gains without a node shrink. The whole reason Apple was “competitive” in the first place was they were on a far superior node.

They don't seem interested in competing or taking marketshare away from Raptor Lake/Zen 4. Perhaps they have hit a frequency wall.

That’s because the two are erm, Apples and Oranges. Different operating systems. The platforms largely don’t compete in most areas.

The main benefit of M2 over M1 is reduced power consumption, which is kind of pointless for the Mini.

Better GPU and CPU performance is needed on the mini.

Apple won’t give the mini higher tier chips because the mini would eat into Mac Studio sales. I really do wish they would at least give us one of the chips beyond the base, however.

 

[FlameTail]

They don't seem interested in competing or taking marketshare away from Raptor Lake/Zen 4. Perhaps they have hit a frequency wall.

Maybe.
Wider designs have lower frequencies. Apple uses very wide designs for their CPU cores.

 

[FlameTail]

Apple won’t give the mini higher tier chips because the mini would eat into Mac Studio sales. I really do wish they would at least give us one of the chips beyond the base, however.

M2 Pro?

 

[Ajay]

M2 Pro?

That's what many of us would like. Seems like a trivial enough upgrade that Apple should just do it; but I have no idea what the actual Mac Mini sales volume is.

Of course, I threaten to move over to MacOS every two years - kind of like the way nuclear fusion is always twenty years away from being a viable power source.

 

[FlameTail]

Apple's M2 Max chip runs Geekbench on a MacBook Pro with 96GB of RAM

The 12-core M2 Max processor adds two extra cores over its predecessor. Apple is testing its upcoming 14-inch and 16-inch MacBook Pros with the as-of-yet...

www.gsmarena.com

M2 Max scores ~14 000 points in Geekbench5 multi. That would put it a notch above 12900K.

Intel Core i9 12900HK vs Apple M2: performance comparison

We compared Intel Core i9 12900HK (2.5 GHz) against Apple M2 (3.5 GHz) in games and benchmarks. Find out which CPU has better performance.

nanoreview.net

It's frankly incredible how powerful Apple's P cores are. 12900HK scores ~1650 while Apple M2 does ~1900, all the while consuming less power and using lower frequency.

 

[igor_kavinski]

Of course, I threaten to move over to MacOS every two years - kind of like the way nuclear fusion is always twenty years away from being a viable power source.

Wish someone would port common apps similar to Notepad/Windows Explorer to the MacOS. Trying to do things the Mac way seems way too constrained.