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

 

[Heartbreaker]

Lenovo on the other hand has no reason other than power saving to solder RAM because it's design is not like Apple's.

LPDDR is typically also faster than regular DDR (at officially supported speeds).

So by soldering in you get lower power, faster RAM, that reduces tech support issues, and takes up less space.

That's a lot of Pro's, now the Cons:

Upsetting a vocal minority of nerds on tech forums...

 

[scannall]

Here is a reason why Apple's RAM is soldered. It's unified memory at LPDDR5 at 6400. LPDDR5 only comes soldered. You would need 8 DIMM slots for the M1 Max Macbook to achieve the same pref at higher power consumption. Now you see why Apple does not use DIMMs.

Now the SSD should be socketed because there is no other reason other than profit.

Lenovo on the other hand has no reason other than power saving to solder RAM because it's design is not like Apple's.

Apple's SSD is unique. The SOC is the drive controller, and the SSD is encrypted and keyed to the specific SOC in your machine. So someone can't just pull the drive and get all the information off of it.

 

[Doug S]

Apple's SSD is unique. The SOC is the drive controller, and the SSD is encrypted and keyed to the specific SOC in your machine. So someone can't just pull the drive and get all the information off of it.

And the NAND is socketed, at least in some models. You just can't easily buy a bigger one on the open market, because Apple is the only company with an "SSD" that's not an SSD but simply raw NAND.

 

[Eug]

According to Taiwan’s Commercial Times, TSMC N3 starts mass production next month, but with TSMC previously saying it won’t see actual N3 revenue until 2023.

I guess that probably means N3 Macs at a spring 2023 announcement.

 

[Doug S]

According to Taiwan’s Commercial Times, TSMC N3 starts mass production next month, but with TSMC previously saying it won’t see actual N3 revenue until 2023.

I guess that probably means N3 Macs at a spring 2023 announcement.

The question is, WHAT N3 Macs will they announce? I remain skeptical that they would have M2 made on N5 and M2 Pro/Max made on N3. My guess is the M2 family is complete, and what gets made on N3 is the M3 family which will be Pro/Max only.

It wouldn't make any sense for them to put A15 cores in something shrunk for a newer process when they could have used A16 cores, nor do I think they'd use A16 cores but still call it M2. We'll see.

 

[Eug]

The question is, WHAT N3 Macs will they announce? I remain skeptical that they would have M2 made on N5 and M2 Pro/Max made on N3.

Why?

 

[Doug S]

Why?

Because they have to do a new design for N3 anyway, so why would they use an outdated core? If it was N4 they could do a simple shrink, then it is understandable. Not with a new node.

 

[gdansk]

M2 Pro and M2 Max are just names after all. Doesn't *necessarily* mean it has to be Avalanche/Blizzard combination like M2. But it'd be weird if it wasn't

 

[smalM]

They had A10 on 16FFC and A10X on N10 (but A10X was launched 3 month before A11).

 

[Eug]

They had A10 on 16FFC and A10X on N10 (but A10X was launched 3 month before A11).

A9 was both Samsung 14 nm and TSMC 16 nm.
A5 was both Samsung 45 nm and Samsung 32 nm.

 

[Eug]

Because they have to do a new design for N3 anyway, so why would they use an outdated core? If it was N4 they could do a simple shrink, then it is understandable. Not with a new node.

Then what will they be making with that N3 capacity for the next 6 months? Certainly not just the Mac Pro SoC since that is too low volume. It also won’t be A16 since N3 is too late for A16. And it won’t be A17 yet either since 2022 Q3 to 2023 Q1 is too early for A17.

 

[Doug S]

Then what will they be making with that N3 capacity for the next 6 months? Certainly not just the Mac Pro SoC since that is too low volume. It also won’t be A16 since N3 is too late for A16. And it won’t be A17 yet either since 2022 Q3 to 2023 Q1 is too early for A17.

Apple Silicon in all forms is too low volume to make a dent in TSMC's N3 production. If they dropped all older Mac and iPad Pro models and went with N3 based Apple Silicon exclusively they couldn't hit even 10k wpm in H1 2023. Given that they keep older models around at a discount they wouldn't need much more than half that.

That's not Apple's problem though, they aren't responsible for filling TSMC's fabs when they release a new process outside the standard iPhone mass production timeline.

 

[Eug]

Apple Silicon in all forms is too low volume to make a dent in TSMC's N3 production. If they dropped all older Mac and iPad Pro models and went with N3 based Apple Silicon exclusively they couldn't hit even 10k wpm in H1 2023. Given that they keep older models around at a discount they wouldn't need much more than half that.

That's not Apple's problem though, they aren't responsible for filling TSMC's fabs when they release a new process outside the standard iPhone mass production timeline.

1. Original rumours stated that Apple and Intel had bought out basically all of TSMC’s initial N3 production. Not sure what’s happening with Intel now but Apple is still on track.

2. This isn’t about TSMC’s N3 capacity anyway. It’s about what Apple is doing with its N3 capacity allocation to them, in late 2022 and early 2023.

As mentioned, it’s not (just) the Mac Pro because that’s too low volume. It’s likely not A16* (at least for the initial few months of A16) because it’s too late. And it’s not A17 (in 2022) because it’s too early.

So, realistically that leaves M2 Pro and M2 Max/Ultra.

*The only caveat here is that it is rumoured that only iPhone 14 Pro and 14 Pro Max are A16, with iPhone 14 and 14 Plus being A15.

 

[MadRat]

Intel will pay TSMC not to build anything just to keep it tied up. Taxpayer money is acoming and will make up for any losses.

 

[gdansk]

Intel will pay TSMC not to build anything just to keep it tied up. Taxpayer money is acoming and will make up for any losses.

TSMC won't do that as they need to actually use the process to iterate yield improvements, etc.

Intel has many different products they make at TSMC outside of CPU/GPUs. Maybe it is an FPGA. Or AI accelerator. Or a new network adapter.

But either way the start of N3 it isn't well-aligned with Apple's product release cycle.

 

[Doug S]

Intel will pay TSMC not to build anything just to keep it tied up. Taxpayer money is acoming and will make up for any losses.

Sure Intel can buy wafers with the intent to scrap them but what would Intel get out of that? It doesn't suddenly make sense to lose money just because you have other money coming. If you knew I was going to give you a million dollars, would you feel it is a fine idea to pile up a few hundred thousand dollars in cash and light it on fire "because Doug S will make up for my losses"?

 

[Schmide]

Intel will pay TSMC not to build anything just to keep it tied up. Taxpayer money is acoming and will make up for any losses.

They've been bitten by the FTC once before. Once bitten twice shy

 

[smalM]

Apple Silicon in all forms is too low volume to make a dent in TSMC's N3 production.

2. This isn’t about TSMC’s N3 capacity anyway. It’s about what Apple is doing with its N3 capacity allocation to them, in late 2022 and early 2023.

What do you both think initial N3 production volume will be?
This year it's very unlikely TSMC will startup with more than Phase 4. Phase 5 may be scheduled for Q1 next year and Phase 6 for the end of next year or may even slip into 2024.

But either way the start of N3 it isn't well-aligned with Apple's product release cycle.

And TSMC told Apple the N3 scheduling years ago. So may be, just may be, Apple isn't surprised...

 

[moinmoin]

And TSMC told Apple the N3 scheduling years ago. So may be, just may be, Apple isn't surprised...

This is the primary reason why all the talks about delays etc. don't make sense. All of that stuff takes place over several years. Even delays have to be known early on before final plannings if they happen since otherwise the whole business becomes untenable. But I guess to some it may sound better that there's some possible last minute drama.

I guess people are too used to how Intel goes about its so far purely internal business. This won't fly if Intel wants to stand a chance against TSMC in the foundry business.

 

[Doug S]

What do you both think initial N3 production volume will be?

The original plan was risk production volumes of 30k wpm, eventually rising to 105k wpm in the latter stages of mass production. TSMC would of course tailor their actual production to what their customers need, so if (for example) they have a demand for only 15k wpm in the first six months that's what they will be set up to produce. It makes no sense to build a capability for 30k wpm if they know they don't have order volumes to utilize it.

Also, it depends on what you mean by "N3". The first gen N3 will have much lower production volumes than N3E which supposedly is being pulled in and will follow N3 closely. If it getting pulled in to Q2 of next year as rumored it may be used for A17 instead of N3 - which if true would mean N3 is a low volume short lived node. Since N3E improves both power and performance, as well as reduces cost thanks to fewer steps, that's where most customers will want to be and that's where TSMC will focus their resources. Sort of like with N7, where most customers were encouraged to go with N7+ so all new designs after N7+ came out have been made on that.

 

[jpiniero]

Also, it depends on what you mean by "N3". The first gen N3 will have much lower production volumes than N3E which supposedly is being pulled in and will follow N3 closely.

TSMC reiterated the +1 year timeframe for N3E during their earnings report in July.

Also like nobody used 7+.

 

[Eug]

https://twitter.com/x/status/1563200192157224965

Ming Chi Kuo is doubling down on the Q1 2023 TSMC 3 nm ship date.

However, he says the M2 Pro/Max MacBook Pros begin production in Q4 2022, so most likely 5 nm.

 

[Doug S]

Didn't Kuo recently claim that new Macs with M2 Pro chips would be announced at Apple's upcoming event, or was that someone else? If that's true they would presumably ship in October. I figured they'd do a separate event a month later like they have in the past, but it depends on how much they plan to announce on the 7th besides iPhone 14.

If Apple is making M2 Pro on N4/N5, the only thing Apple could be using N3 in the first half of next year would be M3! I mean, I suppose someone could posit they use N3 for M2 Max, but that's hardly going to drive much wafer volume. But following up M2 in less than nine months with M3 would be unexpected too.

 

[Eug]

Didn't Kuo recently claim that new Macs with M2 Pro chips would be announced at Apple's upcoming event, or was that someone else? If that's true they would presumably ship in October. I figured they'd do a separate event a month later like they have in the past, but it depends on how much they plan to announce on the 7th besides iPhone 14.

If Apple is making M2 Pro on N4/N5, the only thing Apple could be using N3 in the first half of next year would be M3! I mean, I suppose someone could posit they use N3 for M2 Max, but that's hardly going to drive much wafer volume. But following up M2 in less than nine months with M3 would be unexpected too.

Presumably M2 Pro and Max would be made at the same time, since they ship in the same machines, and in relatively high volume (for Apple).

 

[Doug S]

Presumably M2 Pro and Max would be made at the same time, since they ship in the same machines, and in relatively high volume (for Apple).

You might be right, but the rumor only mentioned M2 Pro. They could introduce new M2 Pro models and not do M2 Max if they were going to either do a Max (whether M2 or M3) in N3 within six months.