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

 

[Doug S]

Commercial Times claims M2 Pro/Max is on 3 nm, even though some here say the timing is off and that N3 volume would be on N3E.

TSMC has always said N3 would go into mass production in H2 2022, and later updated that to say first production wafers would arrive in early 2023. Apple could announce in November, and ship later in November or in December if TSMC is beating that promise by a bit - or Apple is using some late risk production wafers.

It isn't like these M2 variants would be very high volume compared to plain M2 let alone Axx, so the claims that N3E is where the volume comes (from A17 and the M3 family next fall) remain intact.

 

[Eug]

TSMC has always said N3 would go into mass production in H2 2022, and later updated that to say first production wafers would arrive in early 2023. Apple could announce in November, and ship later in November or in December if TSMC is beating that promise by a bit - or Apple is using some late risk production wafers.

It isn't like these M2 variants would be very high volume compared to plain M2 let alone Axx, so the claims that N3E is where the volume comes (from A17 and the M3 family next fall) remain intact.

I wonder how much volume they may have gotten from risk production.

 

[GC2:CS]

M2 Pro/Max to get LPDDR 5X with 8,5Gbps

Next-Generation MacBook Pros Rumored to Feature 'Very High-Bandwidth' RAM

Apple's next-generation 14-inch and 16-inch MacBook Pro models with M2 Pro and M2 Max chips will be equipped with "very high-bandwidth,...

www.macrumors.com

so we have
A14 - LPDDR 4X | M1 - LPDDR 4X | M1 Pro - LPDDR 5
A15 - LPDDR 4X | M2 - LPDDR 5 | M2 Pro - LPDDR 5X
A16 - LPDDR 5 | M3 - 5X ??? | M3 Pro - ???
A17 - 5X ??? | …. | ….

If true it looks nice in a table. But why that might be is questionable. I do not like that Apple stuck with DR 4X for 5 SoC generations. To my knowledge it does present a bottleneck. If new memory is always with less volts and energy per bit, why they start implementing from the highest end parts not the lowest power ones that would benefit the most ?

 

[Doug S]

M2 Pro/Max to get LPDDR 5X with 8,5Gbps

Next-Generation MacBook Pros Rumored to Feature 'Very High-Bandwidth' RAM

Apple's next-generation 14-inch and 16-inch MacBook Pro models with M2 Pro and M2 Max chips will be equipped with "very high-bandwidth,...

www.macrumors.com

so we have
A14 - LPDDR 4X | M1 - LPDDR 4X | M1 Pro - LPDDR 5
A15 - LPDDR 4X | M2 - LPDDR 5 | M2 Pro - LPDDR 5X
A16 - LPDDR 5 | M3 - 5X ??? | M3 Pro - ???
A17 - 5X ??? | …. | ….

If true it looks nice in a table. But why that might be is questionable. I do not like that Apple stuck with DR 4X for 5 SoC generations. To my knowledge it does present a bottleneck. If new memory is always with less volts and energy per bit, why they start implementing from the highest end parts not the lowest power ones that would benefit the most ?

Newer memory standards cost more and the performance and power benefit for a phone is pretty marginal since Apple's SoCs have such huge caches. I'm not sure how much power benefit there is going 4X -> 5 anyway, usually the big improvements in power draw for LPDDR come with the 'X' variant.

 

[Eug]

TSMC has always said N3 would go into mass production in H2 2022, and later updated that to say first production wafers would arrive in early 2023. Apple could announce in November, and ship later in November or in December if TSMC is beating that promise by a bit - or Apple is using some late risk production wafers.

It isn't like these M2 variants would be very high volume compared to plain M2 let alone Axx, so the claims that N3E is where the volume comes (from A17 and the M3 family next fall) remain intact.

Judging by Tim and Luca’s statements, it appears we probably will not get any new Mac hardware in this quarter.

Tim Cook casts doubt on new M2 MacBook Pros in 2022

Based on comments from Apple CEO Tim Cook during the quarterly earnings report, the odds of a November release for a new 14-inch and 16-inch MacBook Pro have dropped.

appleinsider.com

If we end up getting a spring release, that timing could fit with M2 Pro/Max on N3.

 

[Eug]

Judging by Tim and Luca’s statements, it appears we probably will not get any new Mac hardware in this quarter.

Tim Cook casts doubt on new M2 MacBook Pros in 2022

Based on comments from Apple CEO Tim Cook during the quarterly earnings report, the odds of a November release for a new 14-inch and 16-inch MacBook Pro have dropped.

appleinsider.com

If we end up getting a spring release, that timing could fit with M2 Pro/Max on N3.

It seems that's what Gurman thinks too:

Bloomberg - Are you a robot?

testing of the new MacBook Pros and a new Mac mini—along with development of the first Apple Silicon Mac Pro—has ramped up internally.

The new MacBook Pros will continue to look like the current models, but they’ll trade their M1 Pro and M1 Max chips for the first M2 Pro and M2 Max processors. The M2 Max will go to 12 CPU cores, up from 10, and see its top graphics option move to 38 cores from 32.

A new Mac mini remains in development, and the company continues to test versions with the same M2 chip as the 13-inch MacBook Air and MacBook Pro, as well as an M2 Pro chip, which hikes the CPU and graphics counts. If Apple indeed launches the M2 Pro variation, we can expect the company to probably wind down the still-available Intel model.

Apple had earlier hoped to include the MacBook Pro as part of those releases, but rolling out a new Mac in December wouldn’t make much sense. When I first reported on the new MacBook Pros in June, I said that Apple was aiming for a release between the end of 2022 and early 2023, so I wouldn’t consider a March release to be late by any means.

my belief is that the first non-Intel Mac Pro will have options for 24 and 48 CPU cores and 76 and 152 graphics cores—along with up to 256 gigabytes of memory.

I believe that Apple had originally planned to use a variation of the M1 chip, but at some point made the decision to hold off until versions of the M2 with more CPU and graphics cores are available.
That seems like the right decision. The core M1 architecture is based on the A14 chip from 2020, while the basis of the M2 is quite a bit newer. Apple also may be waiting until it can build chips using the 3-nanometer process (the first M2 chips and the M1 are based on 5-nanometer technology). So it’s probably worth the wait.

So, predictions:

Late calendar 2023 Q1 to early Q2 release, but not necessarily for all of these at the same time:

M2 remains on TSMC N5
M2 Pro & Max/Ultra/Ultra x2 all on TSMC N3

Mac mini:
M2 8/10-core (CPU 4+4)
?M2 Pro, perhaps a binned variant (CPU 6+2)

14"/16" MacBook Pro:
M2 Pro 10/20-core (CPU 8+2)
M2 Max 12/38-core (CPU 10+2)

Mac Studio:
M2 Max, 12/38-core (CPU 10+2)
M2 Ultra, 24/76-core (CPU 20+4)

Mac Pro:
M2 Ultra 24/76-core (CPU 20+4)
M2 Ultra x2 48/152-core (CPU 40+8)
Up to 384 GB RAM, but no RAM slots
6 PCIe slots, but no 3rd party GPU support

 

[smalM]

M2 Max 12/38-core (CPU 10+2)

Why do you think it will be a 10+2 CPU? That would be 4P + 4P + 2P + 2E clusters - not very likely...

 

[Doug S]

Why do you think it will be a 10+2 CPU? That would be 4P + 4P + 2P + 2E clusters - not very likely...

They already have 2P clusters in iPhone SoCs and 2E clusters in the M1 family so why do you say "not very likely" when both already exist at Apple, just not in the same product at the moment?

 

[JustBry76]

Anyone know if AnandTech has an A16 review coming?

 

[DrMrLordX]

Anyone know if AnandTech has an A16 review coming?

You'll have to wait and see. With Ian and Andreas gone it's hard to tell what they're up to anymore.

 

[Doug S]

You'll have to wait and see. With Ian and Andreas gone it's hard to tell what they're up to anymore.

If Anandtech doesn't, we probably won't see it from anyone and we'll all be the poorer for it.

If I was going to pay for an article, that's the kind of article I'd happily fork over a few bucks for. Whether there would be enough of us weirdos interested in that level of detail to make that commercially viable is another matter, especially with how easy it would be for someone to copy/paste that article elsewhere.

 

[DrMrLordX]

If I was going to pay for an article, that's the kind of article I'd happily fork over a few bucks for. Whether there would be enough of us weirdos interested in that level of detail to make that commercially viable is another matter, especially with how easy it would be for someone to copy/paste that article elsewhere.

It'd be worth more than Charlie's paywalled articles.

 

[BorisTheBlade82]

You'll have to wait and see. With Ian and Andreas gone it's hard to tell what they're up to anymore.

You mean Andrei (Frumusanu) I guess?

 

[DrMrLordX]

You mean Andrei (Frumusanu) I guess?

Er yeah my bad

 

[Doug S]

It'd be worth more than Charlie's paywalled articles.

His articles are targeted more IT decision makers and Wall Street, not individuals - that's a smaller audience but one willing to pay more. We might be curious to know what he knows about upcoming roadmaps or problems at a company or whatever but when he can sell that stuff on a subscription for hundreds or thousands (I don't really know what he charges) a year then it obviously going to remain out of our reach.

Its a different market than doing highly technical deep dives in a smartphone SoC. Its just unfortunate that everyone else thinks it is good enough to run a few random benchmarks and call it a day without investigating why this one thing is 40% faster while this other thing is only 4% faster and attempting to tease out details of the architecture. Of course Apple could just present at ISSCC and tell us all this but now I'm just dreaming!

 

[DrMrLordX]

Of course Apple could just present at ISSCC and tell us all this but now I'm just dreaming!

Sadly, transparency is bad business to Apple. Or so they seem to believe anyway.

 

[richardskrad]

-

 

[Doug S]

Sadly, transparency is bad business to Apple. Or so they seem to believe anyway.

Part of it may be fear of lawsuits. The more details you make public the more chance someone's lawyer (especially a patent troll's lawyer) will decide that's something they have a patent on, or is at least close enough to try to make that claim in court. Some of it can be figured out via reverse engineering, but some of it may not be possible to know unless Apple states it outright - and that's the best possible evidence you can get in court. Much better than "our expert reverse engineered this and believes the following".

Apple has a cross licensing deal with Intel, but not so sure about AMD or IBM, though the trolls are by far the biggest worry. With a market cap around 20x higher than Intel, AMD or IBM, they have the biggest possible target on their back.

 

[richardskrad]

The M1 chip is 2 years old today. It’s crazy that Intel and AMD are still unable to answer its performance per watt.

 

[Markfw]

The M1 chip is 2 years old today. It’s crazy that Intel and AMD are still unable to answer its performance per watt.

I don't think you understand that their goals are much different. And its possible that some of the 6 series chips (like the 6800U) when undervolted could actually beat the M1. But its for a different market.

 

[A///]

I would not expect any significant hardware changes until apple's "developer" conference in June or their fall release event. It'll be a long time before intel or amd can stride with apple at ppw, but I wouldn't count on apple too much. there's been some breakdown videos on the internet and a lot of the performance by the mac studio and the other device with the glued chips coming down to the accelerators on the chips themselves. It won't be long until amd or intel go for low hanging fruit and stick dual encoders and decoders or more and whatever they can at a cheap enough cost and transistor budget to make life easier for productivity. there's been an unsubstantiated rumor for a while that apple has hit a performance wall with newer internal designs but there's so little evidence of this I couldn't even call it unsubstantiated. it would be interesting to see how zen 5 and intel pan out in 2024 and beyond with their next gen hardware.

the rumor makes as much sense as intel having designed their heterogeneous architecture for apple, when apple was not a prominent client for Intel. then you have that arm news about gpus needing to be arm's in house designs. there's always a good chance apple will go back to intel or even amd in the future. nothing is set in stone.

 

[Doug S]

The M1 chip is 2 years old today. It’s crazy that Intel and AMD are still unable to answer its performance per watt.

Are they trying to? Seems they are much more focused on upping TDPs to 250W, 350W, or more so they can stuff in more cores on the high end, because that's where their best profit per wafer is.

Mac sales have nearly doubled during this period so obviously there's a demand for e.g. high performance fanless laptops in the Apple world, but the fractured nature of the Windows market makes a similar CPU a tougher sell for Intel or AMD. Because if one introduces it and it gains traction the other will soon follow and reduce their revenue by splitting the market.

OEMs have that problem in spades, a Dell or whoever that introduced that product would get some first mover advantage but if it sells well the other OEMs would follow. If it really is successful then the white label Chinese OEMs follow with retail prices equal to Dell/HP/etc. wholesale price which kills your non US market (US customers may scoff at the idea of buying Chinese branded PCs, but those in the EU and rest of world don't share that bias) And all the while those silent laptops are competing against cheaper ones that cost half as much with the same performance, if you're willing to listen to a fan, have less battery life, lower build quality, etc. etc.

I'm not so sure Intel and AMD can't match M1. It is more that the nature of the PC market would make it very difficult for anyone but Apple to create such a market niche.

 

[richardskrad]

—

Intel and AMD can’t make a chip that has the performance and battery life in such a thin and fan-less enclosure as the M2 MacBook Air. Until that day comes, Apple Silicon is the most advanced SoC on earth in my mind.

The only one that can challenge Apple is Qualcomm and who knows when they’ll release a competitor for Windows OEM’s. I have no faith in AMD and Intel. They don’t seem to care that they’re chips are becoming more and more inefficient.

 

[Roland00Address]

Intel and AMD can’t make a chip that has the performance and battery life in such a thin and fan-less enclosure as the M2 MacBook Air. Until that day comes, Apple Silicon is the most advanced SoC on earth in my mind.

The only one that can challenge Apple is Qualcomm and who knows when they’ll release a competitor for Windows OEM’s. I have no faith in AMD and Intel. They don’t seem to care that they’re chips are becoming more and more inefficient.

Intel and AMD can make a chip that has those goals, but those goals will not be actualised without the right OS and software using dedicated specialized silicon for specific tasks. And if the OS and software will not benefit, than you can not sell a third party to buy this “extra” chip with all these unnecessary add ons, for it causes silicon bloat with larger die sizes.

Thus you have something similar (metaphor time) to the printing press where you need a generalised text that everyone wants, with the right amount of complexity but not too complex. Furthermore the availability of this text which can now be made cheaply does not truly spread until you have high literature rates. Thus we do not see the explosion of reading except in a general location, and it took 100 years until the printing revolution to take off after the 1450s.

My point here is that it is not just an engineering challenge, it is a revolution in how computing occurs, but you need a miss mash, a mix and match of apple having some control but not too much control to make it work. Likewise the other strategy Intel and AMD pursued in the 1980s to 2010s totally made sense at the time, but we are now in different markets.

 

[richardskrad]

.

Apple showed with the M1 Ultra that they can match 80-90% of the performance of an Intel 12900K while consuming 35% of the power. Sure, the average Windows user might not be as sensitive to performance per watt as an Apple user but I don’t think Intel and AMD can match the performance per watt of ARM chips, even if they put a ton more resources into it and Windows did arhictectural changes to Windows. I think it comes down to ARM phasing out x86.