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

 

[Eug]

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.

The Ming-Chi Kuo twitter post I linked was just talking about manufacturing for the MacBook Pros, not any specific SoC per se. Or are you talking about the older DigiTimes rumour that only mentioned M2 Pro?

What Kuo was saying was that info suggests the new MacBook Pros are starting production very soon. However, according to TSMC, 3 nm isn't generating revenue until Q1 2023, so he thinks the new MacBook Pros will likely have 5 nm chips. He didn't actually mention either the M2 Pro or M2 Max by name.

BTW, I realize Kuo totally contradicts my own guess from a few posts earlier. I was guessing they were going to put N3 M2 Pro and M2 Max in the new MacBook Pros, but not until 2023.

If that's wrong and the new MacBook Pros are indeed N5, then what are the new N3 chips in Q1 2023? M3 and A17? For M3, that would mean a pretty short update interval, since M2 was just released 2 months ago. For A17, that seems a tad early too. Or it could be something completely different.

Yes, a new chip for the Mac Pro might be N3, but that would be very low volume. Speaking of the Mac Pro, I'm still wondering if it might have a different naming scheme, to reflect a different memory architecture. However, others seem to think it could be a Max x 4 / Ultra x 2 variant with the same memory architecture. If so, will the Mac Pro be limited to 256 GB RAM then? How many people will actually care if a Mac Pro can't handle 2 TB of RAM?

 

[gdansk]

There were other rumors that N3 was relegated to low volume products and N3E was the new N3 because it was doing better.

 

[Doug S]

There were other rumors that N3 was relegated to low volume products and N3E was the new N3 because it was doing better.

N3 can only be relegated to low volume products if 1) Apple does M2 Pro / M2 Max on N4/N5 like M2; 2) M3 will follow far enough behind it is on N3E; 3) N3E reaches mass production in time to make A17s and hit Apple September iPhone 15 schedule.

If they don't have any high volume wafer needs for N3 they would presumably accelerate N3E as much as possible, to avoid having the equipment on the N3 line sitting idle instead of generating revenue.

 

[Eug]

There were other rumors that N3 was relegated to low volume products and N3E was the new N3 because it was doing better.

Was this already posted? TSMC has 7 major customers lined up for N3

TSMC is only expected to start the production with a mere 1,000 wafer starts a month, which seems like a very low figure, especially as this is said to remain unchanged through all of Q4. On the plus side, yields are expected to be better than the initial 5 nm node yields. Full-on mass production for the 3 nm node isn't expected to happen until the second half of 2023 and TSMC will also kick off its N3E node sometime in 2023.

Apart from Apple, major customers for the 3 nm node include AMD, Broadcom, Intel, MediaTek, NVIDIA and Qualcomm.

I guess that means N3 initially just for the Mac Pro could make sense.

 

[Doug S]

The article only says they have 7 major customers "for the 3nm" node, not that any of those customers will be necessarily be taking any N3 wafers. All but one could be waiting until sometime in 2023, or waiting for N3E, but any one of those customers could easily consume more than 1000 wpm if they were all in. If the yields are fine what's the issue with N3, does it not reach power, performance, or density metrics? Does it cost too much?

Guess that terrible fab utilization is the price they pay for not lining up with Apple's schedule.

 

[MadRat]

If Apple paid for exclusive access for the first year, would that be illegal in any way?

 

[gdansk]

If Apple paid for exclusive access for the first year, would that be illegal in any way?

Not in the US or Taiwan. So no.
But TSMC should avoid such arrangements unless in desperate need of capital. It often ends poorly for suppliers that are beholden to Apple.

 

[poke01]

Apple

It often ends poorly for suppliers that are beholden to Apple.

Apple and TSMC have a special relationship more so than AMD does. The iPhone enabled Apple to fund it's ARM CPUs and TSMC benefited from Apples iPhones orders a lot and made a lot of revenue.

I don't think Apple would do TSMC bad and vice versa. They both need each other.

 

[Doug S]

Not in the US or Taiwan. So no.
But TSMC should avoid such arrangements unless in desperate need of capital. It often ends poorly for suppliers that are beholden to Apple.

Why? If the payments were equal to buying up all the wafers they could produce that would be a GREAT deal for TSMC - they would get the revenue for say 30k wpm while incurring the costs for manufacturing only 1k wpm. They'd still pay the full deprecation on their fabs, but save a ton on per wafer costs. That would be like if you're a pizza joint owner and you have someone paying to buy up all your capacity for making pizzas while asking for only a few dozen a week to actually be made. You would save on all the food costs for the pizzas you don't make, be able to turn off the ovens and air conditioning other than than few hours a week you're open, and save money not having to pay your delivery guys etc. Same income, lower costs - any business would line up for a deal like that!

There's absolutely zero chance Apple would make such a deal though. If they did it would be contingent on the process arriving on their schedule, not arriving halfway between iPhone launch cycles. Even if they did due to some imagined reasons they don't want to compete with e.g. Qualcomm they could sell that capacity to someone who doesn't compete with them at all like Broadcom or Nvidia.

 

[Mopetar]

Not in the US or Taiwan. So no.
But TSMC should avoid such arrangements unless in desperate need of capital. It often ends poorly for suppliers that are beholden to Apple.

If Apple will pay more than anyone else for exclusive access I fail to see how it hurts TSMC. A fab isn't anything like a component vendor. Apple can easily swap out one vendors flash memory for another, but good luck finding someone else who can take the masks for your new SoC and produce the 10's of thousands of wafers per month Apple needs for their newest phone and tablet models.

The only way it hurts TSMC is if they could produce more than Apple wants to buy and aren't realizing any revenue from that excess capacity. It seems unlikely that they'd do something like that and if Apple didn't want to buy the full capacity at the onset, they probably wouldn't care if it were sold to a company like AMD or Intel who aren't making chips for mobile phones.

 

[gdansk]

If Apple will pay more than anyone else for exclusive access I fail to see how it hurts TSMC. A fab isn't anything like a component vendor. Apple can easily swap out one vendors flash memory for another, but good luck finding someone else who can take the masks for your new SoC and produce the 10's of thousands of wafers per month Apple needs for their newest phone and tablet models.

The only way it hurts TSMC is if they could produce more than Apple wants to buy and aren't realizing any revenue from that excess capacity. It seems unlikely that they'd do something like that and if Apple didn't want to buy the full capacity at the onset, they probably wouldn't care if it were sold to a company like AMD or Intel who aren't making chips for mobile phones.

Apparently Apple doesn't seem to buying much of N3. And no one else has designs ready because Apple was supposed to be the main launch customer. If these are true then it is already backfiring.

Apple is the customer where missing a quarter means losings billions. It's high risk, high reward.

 

[poke01]

It's not that Apple is not using TSMC 3nm but rather TSMC missed the deadline for it be ready in time for iphone. Was it ever going to ready this year? Wei said 3nm was difficult.

Majority of rumours said 3nm will be used next year with A17.

 

[jpiniero]

If the yields are fine what's the issue with N3, does it not reach power, performance, or density metrics? Does it cost too much?

I think it is a cost too much problem. The N4 variants are similar in power and performance and way cheaper.

Majority of rumours said 3nm will be used next year with A17.

Doesn't appear to be the case. Looks like it's using N4 (ish).

 

[smalM]

And you know that because...

 

[Eug]

iPhone 14 to get enhanced A15 according to WSJ

As for the base iPhone 14, it will get an “enhanced version” of the existing A15 chip, says unnamed sources.

---

For reference:

iPhone 13/mini:

CPU	Hexa-core (2x "high-performance" 3.23 GHz Avalanche + 4x "energy-saving" 2.02 GHz Blizzard)
GPU	Apple-designed 4 core

iPhone 13 Pro/Max:

CPU	Hexa-core (2× "high-performance" 3.23 GHz Avalanche + 4× "energy-saving" 2.02 GHz Blizzard)
GPU	Apple-designed 5 core

Maybe the iPhone 14 will just get what's already in the iPhone 13 Pro? That would make sense since the A15 in the iPhone 13/mini is just a binned version. Presumably yields are good enough now so the iPhone 14/Plus won't need the 4-core binned GPU version.

 

[Doug S]

Can't be much of an "enhancement" if they haven't increased the clock at all, despite a year's worth of maturity for the process and Apple's conservative frequency binning strategy. So yeah probably just the 5th core.

I'm kind of surprised we haven't seen any credible A16 performance rumors yet, though I'm not expecting much since N4 doesn't really buy much frequency increase or transistor budget. Also they've likely been devoting their top resources to all the fancy interconnect and packaging in the AS line, which isn't done until the Mac Pro is ready to ship, so I'm expecting another single digit IPC boost.

 

[ashFTW]

Also they've likely been devoting their top resources to all the fancy interconnect and packaging in the AS line … Mac Pro …

That seems reasonable, but the AR/VR products which will likely have different/tweaked AS is a much higher priority for Apple hardware group, compared to Mac Pro.

 

[poke01]

so I'm expecting another single digit IPC boost.

yep A16 is 1-2% IPC boost, clocks speed is 3.5Ghz. Single core score is same as M2. New IP and ARMv9 won't come till A17 I think.

 

[Doug S]

What I still don't understand is why they are recycling the A15 for the non-Pro iPhones at all. I was even skeptical at first that rumor was true since they've never done anything like that before.

If the A16 was made on a more expensive process, like N3, then it would be understandable why they might not want to use it across the whole line. But N4 is cheaper than N5P per transistor, so unless A16 is a lot bigger the cost should be similar. It seems like it would make more sense to bin the chips beyond a single working GPU core and use the ones with better frequency/power curves in the Pro phones - so they would have higher performance, or use less power to perform the same fixed amount of work.

By doing it this way they are effectively losing the ability to bin the A16 like they did the A15 last year, unless the Pro Max gets an GPU core over the Pro. Which would make some sense I suppose, since it has more pixels to push. Its either that, or they would have to warehouse tens of millions of A16s that don't qualify for iPhone 14 Pro for a year to use in the iPhone 15 non-Pro.

It will be interesting to see how they frame it in the event tomorrow, and what they choose to focus on as the advantages of the A16 in the 'Pro' phones.

 

[Mopetar]

They always have the option of ignoring the CPU/GPU performance and talking up improvements to other parts of the silicon, or just focusing on camera, screen, etc. improvements.

 

[Doug S]

They always have the option of ignoring the CPU/GPU performance and talking up improvements to other parts of the silicon, or just focusing on camera, screen, etc. improvements.

True, but I expect if they're going to split between A15 and A16 they will tout something about the A16 as an advantage of the Pro line. If they aren't going to mention it, and it costs them the same either way, why bother making that a difference between the two lines?

 

[Eug]

True, but I expect if they're going to split between A15 and A16 they will tout something about the A16 as an advantage of the Pro line. If they aren't going to mention it, and it costs them the same either way, why bother making that a difference between the two lines?

I'm not sure if this is what you're getting at, but I suspect the 48MP / 8K camera requires A16.

Just take a look at A15's press blurb:

 

[Doug S]

I was referring to what they will call out specifically and highlight in the event tomorrow. What they spend more time than just flashing a card like the above on the screen during the event and spend more than a sentence or two talking about. What are they going to tout as a reason why you want to buy the Pro and get the A16? If the only advantage of the A16 is "a little bit faster and handles 8K video" that's not a reason to want the A16 specifically.

And more to the point, wouldn't explain why the non-Pro iPhones weren't getting the A16. I figure it pretty much has to be significantly larger - at least 25% larger - than the A15 for there to be a real cost driver as a reason.

I've seen several rumors indicating the Pros will get a price increase but the base iPhone 14 will remain at the same $799. Now that might have to do with the "4th model" being the Max which is above the base iPhone 14 whereas last year the 4th model was the $699 Mini below the base iPhone 14. Maybe only the Pro gets the rumored satellite capability, and if it offers free texts (though maybe only in an emergency?) they'd need to build in some cost for that service I suppose.

Oh well, enough speculation I guess since we'll find out tomorrow. I just find a lot of this weird, and have a feeling I will still find it weird after...and maybe even after we get some teardowns and find out something about the A16's size and layout.

 

[Eug]

Yep. iPhone 14 has A15 with 5-core GPU. Seems to be same chip as iPhone 13 Pro.

 

[jpiniero]

What I still don't understand is why they are recycling the A15 for the non-Pro iPhones at all. I was even skeptical at first that rumor was true since they've never done anything like that before.

I believe it's a combination of the A16 being more expensive and really wanting to push the Pro models.