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

 

[FlameTail]

He got a new subscriber, really impressed by this video.

I’m pretty convinced that TSMC N3 got shipped prematurely. I get why Apple was allegedly paying for known good die now, the economics of N3B wouldn’t make sense otherwise. The N3B node seems to be a practical joke.

Now it all makes sense why Mediatek and Qualcomm are sticking to 4nm for their next generation SoCs : Dimensity 9300 and SD8G3.

 

[FlameTail]

He got a new subscriber, really impressed by this video.

Yup. I guess we finally have a true succesor to Andrei's in-depth reviews of mobile CPUs at Anandtech.

 

[jeanlain]

Impressive.

Widest core in the industry, eh?

The CPU everyone has noticed on this chart is not the one that the poster pointed out. There, huh, a phone CPU in there.

 

[igor_kavinski]

World's most expensive handheld console:

I bet Apple is really proud of doing the Robinhood thing of taking money away from the filthy rich.

I dare you not to quote me and ask me, "what would any other company have done?"

 

[SpudLobby]

I have said for a couple years that M3 would use A17 cores. If M3 waits a bit longer than I guessed (I was always assuming fall 2023) for N3E which seems more and more likely then IMHO there's a good possibility that even the A17s made on N3B are actually hybrid (pessimistic) N3B/N3E designs - i.e. they won't take advantage of the (slightly) denser cache possible with N3B, nor will they take advantage of the features that N3E offers over N3B. We'll know if this is true if the cache does not shrink at all from A16's when Techinsights does one of their die evaluations.

Then they wouldn't need new mask sets or changes to the design, TSMC would keep using the same masks on N3E once that's ready and the only way you might be able to tell is if you run some sort of CPU burn test designed to drain the battery which would last a bit longer on iPhone 15s with N3E A17s. The cost of that assuming the cores are basically ported over would be M3s would underperform a bit versus what they are capable of, but they should be enough of a jump over M2 it doesn't matter (and that makes it easier for M4 to exceed as well, despite only getting a bump to N3P)

I don't think this is true at all RE: how easy it would be to port but it's nice you're now coming around to the idea A17 on N3E in time, and no I don't think it's N3E currently. They aren't design rule compatible, it just not some hybrid process, it's N3B.

N3E ramp is H2 2023 sure but that's not soon enough. Porting it wouldn't necessarily be a huge cost for next year but I do not in fact think this is a "hybrid N3E/N3B design" or that the M3 could use N3E. It does not work like that. The differences between N3E and N3B extend beyond the cache size.

At any rate the rumors about the M3 are that it's delayed, the chips are done and it's primarily a function of N3B supply which also makes sense given the parametric and/or catastrophic yield issues.

 

[SpudLobby]

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")

It's interesting they called the A17 the A17 Pro. I think this points to the idea they might have a lower clocked A17 standard that's coming to the iPhone 16. We could say it's all just branding and while that's possible, given their current strategy that slates the current year chip into last year's phone, it does not make sense to have the A17 "Pro" in an iPhone 16 standard.


Certainly looks to me like we might see an A17 standard on N3E, probably with lower clocks and a smaller GPU and that N3B yields aren't great (though Apple probably screwed up some arch too, but). No way they'll pass the A17 to the iPhone 16 under the A17 Pro banner.

 

[Eug]

One thing I haven't seen discussion of that probably deserves it is the name. Apple specifically chose to name it "A17 Pro".

That kind of implies to me that the non-Pro iPhone 16s will not get this SoC next year. If not, what will they get? Will it be as simple as one less GPU core (which they have binned on already) or will there be larger differences between A17 Pro and next year's "A17 Fusion" or whatever it may be called in the iPhone 16? Is it possible it will be a separate SoC design, and if so what might they differentiate on?

It's interesting they called the A17 the A17 Pro. I think this points to the idea they might have a lower clocked A17 standard that's coming to the iPhone 16. We could say it's all just branding and while that's possible, given their current strategy that slates the current year chip into last year's phone, it does not make sense to have the A17 "Pro" in an iPhone 16 standard.


Certainly looks to me like we might see an A17 standard on N3E, probably with lower clocks and a smaller GPU and that N3B yields aren't great (though Apple probably screwed up some arch too, but). No way they'll pass the A17 to the iPhone 16 under the A17 Pro banner.

Well, we have had the A10 Fusion and a bunch of Ax Bionic chips too. I’ve never understood the point of these suffixes. So, while you could be right, I’m not yet totally convinced. And yes I could see the “Pro” chip in the lower end models… a year later.

That said, that could explain the A17 N3B / N3E rumour. Maybe A17 Pro is N3B, and then in 2024 there will be the A17 on N3E for iPhone 16 and 16 Plus. It would be a different chip, based on the A17 core, but not a binned A17 Pro.

 

[Henry swagger]

Impressive.

Widest core in the industry, eh?

I think the x4 arm will have a 11 wide decode

 

[Ajay]

I don't think this is true at all RE: how easy it would be to port but it's nice you're now coming around to the idea A17 on N3E in time, and no I don't think it's N3E currently. They aren't design rule compatible, it just not some hybrid process, it's N3B.

N3E ramp is H2 2023 sure but that's not soon enough. Porting it wouldn't necessarily be a huge cost for next year but I do not in fact think this is a "hybrid N3E/N3B design" or that the M3 could use N3E. It does not work like that. The differences between N3E and N3B extend beyond the cache size.

At any rate the rumors about the M3 are that it's delayed, the chips are done and it's primarily a function of N3B supply which also makes sense given the parametric and/or catastrophic yield issues.

Yeah, having to deal with a new PDK is a major problem. I image all the RTL characteristics are different and all macros will need to be redone. If that's the case, I would think the expected launch of M3 Macs would be pushed out 6-9 months from the original target date (architecture would still be the same). N3E would be a much better choice, particularly for laptops, but Apple may just dance with the one they brought to the party (N3b) as a release timing issue.

 

[Panino Manino]

I think the x4 arm will have a 11 wide decode

Is this really necessary? Makes me think about how Zen 4 can compete with Intel while being narrower.

 

[SpudLobby]

Yeah, having to deal with a new PDK is a major problem. I image all the RTL characteristics are different and all macros will need to be redone. If that's the case, I would think the expected launch of M3 Macs would be pushed out 6-9 months from the original target date (architecture would still be the same). N3E would be a much better choice, particularly for laptops, but Apple may just dance with the one they brought to the party (N3b) as a release timing issue.

Yeah. I think they’ll dance with it but given the Mac’s scale it’s possible that they decide it’s better to eat the upfront cost and extra time because they can’t get capacity on N3B early enough anyways which is the current rumor and might be confused for the former. It depends on how bad the N3B parametric yields and cost really are IMO.

 

[Doug S]

M3 looks really promising. Question is when will they release it. Since that has way more of performance cores, TDP will have to go up for sure relative to M2. But these chips are so efficient relative to x86 chips that it does not matter at all for laptop form factor. I only hope base RAM goes up from 8GB(I am not too optimistic)

I'm still betting on them releasing it next year so it is on N3E. That should address the power issues that A17 appears to have. Avoiding even a small increase in TDP is important to Apple for laptops because they have some models that don't include a fan at all.

Also echo the other post that it will be 12 GB base this time around.

 

[Mopetar]

They might put out the regular M3 on N3B if there's room for it. It really depends on the yield as well as ensuring enough iPhone SoCs can be made.

A lot of their products that would use the PRO/MAX/ULTRA chips don't need a refresh so they can easily hold out for N3E for those products.

I'm assuming that A17 went into production FSR enough ahead of M3 that they would have been able to make some design corrections for using N3B. N3E seems far enough out that they would have had to commit to N3B for the M3 to be able to release any new Macs this year.

 

[trivik12]

N3E for M3 is unrealistic unless they delay it to H2 2024. It takes a while for it to ramp up after mass production begins and its supposed to do that in H2 2023(similar to N3B in H2 2022). That generally means end of the year and then it takes time for yields to go up. It will debut with A18 pro (or whatever they call the soc on iphone 16 pro/ultra). I think we will see new macs by March next year.

 

[Ajay]

N3E for M3 is unrealistic unless they delay it to H2 2024. It takes a while for it to ramp up after mass production begins and its supposed to do that in H2 2023(similar to N3B in H2 2022). That generally means end of the year and then it takes time for yields to go up. It will debut with A18 pro (or whatever they call the soc on iphone 16 pro/ultra). I think we will see new macs by March next year.

N3E goes HVM This year. https://www.anandtech.com/show/1883...n-schedule-n3p-n3x-deliver-five-percent-gains

So, yeah, M3 could be on N3E and in MACs first half of next year.

 

[Doug S]

N3E for M3 is unrealistic unless they delay it to H2 2024. It takes a while for it to ramp up after mass production begins and its supposed to do that in H2 2023(similar to N3B in H2 2022). That generally means end of the year and then it takes time for yields to go up. It will debut with A18 pro (or whatever they call the soc on iphone 16 pro/ultra). I think we will see new macs by March next year.

TSMC said N3E has hit all performance and yield targets and will enter mass production in H2 2023. Even if that means "December" there should be finished wafers rolling out in March/April. If mass production starts earlier than December then finished wafers are available earlier in the year.'

That's why I suggested N3E based M3 Macs "next spring" which would cover even the worst case of mass production starting on December 31st.

 

[varase]

How did you choose those score numbers ?

I'm getting

Single 2,963
Multi 7,355
Metal 27,088

on my 15 Pro Max.

When I saw those AAA games running, I decided to get it with 1 TB. 😄

Of course, this is with all my stuff loaded, and battery is still dropping so I believe it's still indexing - probably more intensive now that iOS can now extract text from photos and videos.

Anyone know if that text gets included in the Spotlight index?

Got about 316 GB loaded including TV shows, movies, stuff I've shot, podcasts, music, documents, and books (including PDFs).

 

[varase]

So does Apple A17 use ARMv9 now?

I don't think so … Apple Silicon contains a lot of what ARMv9 tried to accomplish anyway.

In fact, from what I understand a lot of ARMv9 is an attempt to catch up to Apple Silicon.

I've heard they increased A17 to nine wide as opposed to eight wide - can't believe they even made the attempt considering how often eight wide would probably get triggered.

Anyone know if they're planning a deep dive into A17? I don't even know what the core names are …

 

[FlameTail]

Anyone know if they're planning a deep dive into A17? I don't even know what the core names are …

They?

Who?

 

[Doug S]

I've heard they increased A17 to nine wide as opposed to eight wide - can't believe they even made the attempt considering how often eight wide would probably get triggered.

I think it was more the reorganization of what integer instructions can slot into which queues and dispatch buffers that was the driver for that. They went from 6 to 8 integer ALUs, if you add in the 4 load/store units and 4 FP units they go from 14 to 16 execution units in total. So probably a bump in the decode was warranted. No one is expecting 8 or 9 wide to happen very often, but if you look at Geekerwan's block diagram (still want to know how he got that...it would have to come from inside Apple because there's no way reverse engineering can get that level of detail) there are some more important differences between A16 & A17.

Look at the dispatch buffers for the ALUs, in the A16 they have dedicated schedulers for each of the six ALU units. In the A17 not only are there now 8 ALU units, but there is some sharing of schedulers among the similar ALU types which should allow for greater efficiency. Ditto for the middle group of integer ALUs - the ones that cannot execute either branches or implement more complex instructions like integer multiply and divide.

Changes like this are made after extensive simulation of real world code, and may go hand in hand with compiler changes that reorder instructions and allow the A17's updated design to relieve some bottlenecks they've observed in those simulations. It is quite possible these changes actually had little to do with the iPhone's needs and were targeted at the Mac. Since both products share the same cores, if they want to make changes that are intended to benefit the Mac they come along for the ride on the iPhone even if they might make little difference. Look at the load/store scheduler and queue sizes between A16 and A17, for example. I have to think that's something that will make a far bigger impact on the Mac (for certain application types) but would never be noticeable for typical phone software loads.

 

[eek2121]

I'm getting

Single 2,963
Multi 7,355
Metal 27,088

on my 15 Pro Max.

When I saw those AAA games running, I decided to get it with 1 TB. 😄

Of course, this is with all my stuff loaded, and battery is still dropping so I believe it's still indexing - probably more intensive now that iOS can now extract text from photos and videos.

Anyone know if that text gets included in the Spotlight index?

Got about 316 GB loaded including TV shows, movies, stuff I've shot, podcasts, music, documents, and books (including PDFs).

For reference, my heavily used 14 Pro Max gets 2647/7044.

 

[Eug]

I'm getting

Single 2,963
Multi 7,355
Metal 27,088

on my 15 Pro Max.

When I saw those AAA games running, I decided to get it with 1 TB. 😄

Of course, this is with all my stuff loaded, and battery is still dropping so I believe it's still indexing - probably more intensive now that iOS can now extract text from photos and videos.

3000 SC achievement unlocked!

These are recent Geekbenches:

iPhone16,1 - Geekbench

Benchmark results for an iPhone16,1 with an ARM processor.

browser.geekbench.com

iPhone16,2 - Geekbench

Benchmark results for an iPhone16,2 with an ARM processor.

browser.geekbench.com

SC - 3002
MC - 7823
Metal - 27526

Interestingly, this twitter post from March was basically bang-on:

https://twitter.com/x/status/1635500693531525120

SC - 3019
MC - 7860

 

[jpiniero]

^^^^

Did they test it in the Freezer?

 

[Eug]

^^^^

Did they test it in the Freezer?

Actually, I was thinking just that.

 

[Mopetar]

^^^^

Did they test it in the Freezer?

Dunno if that's cold enough. Might have had to use liquid nitrogen to pull that off.