Discussion Apple Silicon SoC thread

Page 172 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Eug

Lifer
Mar 11, 2000
23,583
996
126
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:

Screen-Shot-2021-10-18-at-1.20.47-PM.jpg

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:


M1 Ultra discussion here:


M2 discussion here:


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:

 
Last edited:

Eug

Lifer
Mar 11, 2000
23,583
996
126
JIT Inventory, doesn't mean they didn't sign long term contracts that obligate them to buy a certain amount.

It's a near certainty that they signed big long term contracts to get good volume discounts.
You realize of course they would want access to Intel chips long term in order to be actually able to provide Intel chips long term.

It would have been unthinkable to drop all Intel machines in 2020. My original prediction was they would probably drop them in 2022. It turns out it is January 2023 for the Mac mini. That doesn't seem too far off, but makes sense since they didn't release the replacement Mac mini until 2023. And it looks like it will likely be 2023 for the Mac Pro as well.
 

Heartbreaker

Diamond Member
Apr 3, 2006
4,222
5,224
136
It would have been unthinkable to drop all Intel machines in 2020.

Well of course, since they didn't have replacements for most of the lineup. Naturally it was expected that they would be dropped when they had a corresponding ARM replacement, and it was planned for a 2 year replacement schedule.
 

repoman27

Senior member
Dec 17, 2018
342
488
136
The M2 Pro Mac mini can go up to 8TB SSD and there is plenty of space for socketed NAND drives. Doing so in a desktop is plain stupid and shows that Apple does not care about data safety at all.

Please don't talk about backups, thats not an excuse to solder the SSD Apple!!.( This is Apples defence)

More e-waste design from Apple cause if the internal SSD dies the Mac is a dead product and NO you can't boot from an external SSD on AS Macs in the future if the internal SSD dies.


Anyone can that can defend this behavior?

Sorry for the rant but great job on the the analysis.
I think the best course of action if Apple wanted to try to make everyone happy would be to keep the integrated storage offerings exactly as they are (barring the outlandish upgrade prices, of course), but also include a bog standard user accessible M.2 slot in every Mac and a dead simple way to point the macOS Data volume to an APFS container on an alternate internal or external drive.

It's important to realize that Apple deliberately architected their solution to fail-secure rather than fail-safe. It's also worth noting that the SSD controller is integrated into the SoC and is designed to tolerate full-die NAND failures. If the SSD were to die in an Apple Silicon Mac, then your CPU, GPU, and RAM are probably all dead as well, and you'll need a new logic board anyway. If any of the NAND packages were to die, a logic board replacement would once again do the trick, although skilled technicians with access to the necessary parts can also replace the NAND modules at the component level. These types of failures appear to be vanishingly rare though, especially seeing as Apple uses the same NVMe storage controller and NAND architecture in iPhones and iPads as well.

The biggest issue I've heard of with putting SSDs down on the logic board is the situation that arises when the controller and NAND are perfectly fine, but you can't get power to those devices due to a component failure somewhere else on the board. This could be alleviated by having a robust USB recovery mode (mostly there), modular USB ports that can be replaced in the event of physical damage (check), and a header on the logic board that technicians could use to connect an external power supply and bypass the low level power management and DC-DC conversion circuitry (no idea if this exists or not).
 

moinmoin

Diamond Member
Jun 1, 2017
4,934
7,620
136
It turns out it is January 2023 for the Mac mini. That doesn't seem too far off, but makes sense since they didn't release the replacement Mac mini until 2023. And it looks like it will likely be 2023 for the Mac Pro as well.
🤔 You make it sound like we aren't in January 2023 yet.
 
  • Haha
Reactions: Eug

Doug S

Platinum Member
Feb 8, 2020
2,205
3,409
136
Just your guess, we have no idea why the Intel Mini lingered.


My guess: people who want to be able to run x86 Windows rather than ARM Windows.

I'm not sure how big the dual boot/VM Windows userbase for Macs is/was, but that userbase is likely nearly 100% business. Corporations would see nothing wrong with buying an Intel Mac in 2022 as a path of least resistance for their Mac users that need to be able to run x86 Windows stuff and don't want the performance hit and uncertainty of Windows/ARM's x86 emulation.
 

Heartbreaker

Diamond Member
Apr 3, 2006
4,222
5,224
136
My guess: people who want to be able to run x86 Windows rather than ARM Windows.

I'm not sure how big the dual boot/VM Windows userbase for Macs is/was, but that userbase is likely nearly 100% business. Corporations would see nothing wrong with buying an Intel Mac in 2022 as a path of least resistance for their Mac users that need to run x86 Windows stuff and don't want the performance hit and uncertainty of Windows/ARM's x86 emulation.

If they need to run Windows, the vast majority are just going to buy a Windows PC.
 

Doug S

Platinum Member
Feb 8, 2020
2,205
3,409
136
If they need to run Windows, the vast majority are just going to buy a Windows PC.


Well looking at market share, that's what the vast majority have always done.

The point is that there are people who want/need a Mac but their corporate overloads require one or more Windows apps they have to access for their job. These are the people who have dual booted or run Windows in a VM traditionally, and there have been enough of them that Apple formally supported this usage.
 

poke01

Senior member
Mar 8, 2022
651
612
106
I think the best course of action if Apple wanted to try to make everyone happy would be to keep the integrated storage offerings exactly as they are (barring the outlandish upgrade prices, of course), but also include a bog standard user accessible M.2 slot in every Mac and a dead simple way to point the macOS Data volume to an APFS container on an alternate internal or external drive.

It's important to realize that Apple deliberately architected their solution to fail-secure rather than fail-safe. It's also worth noting that the SSD controller is integrated into the SoC and is designed to tolerate full-die NAND failures. If the SSD were to die in an Apple Silicon Mac, then your CPU, GPU, and RAM are probably all dead as well, and you'll need a new logic board anyway. If any of the NAND packages were to die, a logic board replacement would once again do the trick, although skilled technicians with access to the necessary parts can also replace the NAND modules at the component level. These types of failures appear to be vanishingly rare though, especially seeing as Apple uses the same NVMe storage controller and NAND architecture in iPhones and iPads as well.

The biggest issue I've heard of with putting SSDs down on the logic board is the situation that arises when the controller and NAND are perfectly fine, but you can't get power to those devices due to a component failure somewhere else on the board. This could be alleviated by having a robust USB recovery mode (mostly there), modular USB ports that can be replaced in the event of physical damage (check), and a header on the logic board that technicians could use to connect an external power supply and bypass the low level power management and DC-DC conversion circuitry (no idea if this exists or not).
They could what do in the Mac Studio and Mac Pro offer socketed NAND drives. This ensures only the NAND is replaced and not the whole board.


Imagine if PC users had the SSD/GPU/CPU soldered to the motherboard and when the SSD died you replaces whole motherboard for just one broken part. That to me is not good design, it's not thoughtful of the users time or money nor the environment which Apple seems to care about on stage.

Apple Sillicon is not designed to flexible for the end user and Apple is proving that with their designs
 

Doug S

Platinum Member
Feb 8, 2020
2,205
3,409
136
People worry way too much about hitting the wear limits on their NAND. I put an SSD in my Tivo and despite 24x7 recording on four tuners (I don't ever put it in standby mode) for two years SMART tells me I have only used 20% of its capacity. Unless you are hammering on your storage to the tune of a hundred MB per second round the clock this isn't something you need to worry about.

The main failure mode for SSDs is electronic - the controller goes bad. Apple's "SSDs" are raw NAND, they don't have a controller - that's built into the SoC. So the only thing that can possibly go wrong with them is wearing out from overuse. Sockets would be nice for upgrades, but pointless for repair.
 

poke01

Senior member
Mar 8, 2022
651
612
106
So the only thing that can possibly go wrong with them is wearing out from overuse. Sockets would be nice for upgrades, but pointless for repair
. No sockets would also be good for quick replacements if something goes wrong. If the NAND dies the whole Mac is rendered unusable and you would have to go Apple for a complete board replacement. Just crazy they would solder NAND on a desktop.

Apple offers socketed NAND on Mac Pro and Studio. For the Studio Apple offers NAND replacements in their self repair program and fact that they do shows that it is useful for repair. It saves time and money.

Apple treats its "normal" consumers differently from their Pro customers and what Apple is saying is that pay extra if you want socketed NAND and even more if you want PCIe.
 

Mopetar

Diamond Member
Jan 31, 2011
7,797
5,899
136
It's a bit of a weird perspective, but at a certain point even expensive equipment can become less expensive to replace outright than to do repair work on. Labor costs are what ultimately drive these decisions.

Apple is certainly angling to soak up repair fees as well and an all-in-one approach makes them really the only game in town for a lot of problems. It also prevents anyone from buying a cheaper model and dropping in a larger drive to avoid the Apple tax.

If you look at it from the perspective of what it costs Apple for an entire board, rather than what they charge for the whole computer, it suddenly becomes clear that it could be more expensive to have a skilled technician spend time diagnosing and fixing a specific problem rather than a less skilled individual who can swap a single component.
 

moinmoin

Diamond Member
Jun 1, 2017
4,934
7,620
136
It's a bit of a weird perspective, but at a certain point even expensive equipment can become less expensive to replace outright than to do repair work on. Labor costs are what ultimately drive these decisions.
Not a weird perspective but a if not the major excuse behind planned obsolescence. "It's not easily repairable since labor costs would be too high to actually repair it." "Let's make it even harder to repair it due to that to save more money." Loop back again.
 

Doug S

Platinum Member
Feb 8, 2020
2,205
3,409
136
It's a bit of a weird perspective, but at a certain point even expensive equipment can become less expensive to replace outright than to do repair work on. Labor costs are what ultimately drive these decisions.

Apple is certainly angling to soak up repair fees as well and an all-in-one approach makes them really the only game in town for a lot of problems. It also prevents anyone from buying a cheaper model and dropping in a larger drive to avoid the Apple tax.

If you look at it from the perspective of what it costs Apple for an entire board, rather than what they charge for the whole computer, it suddenly becomes clear that it could be more expensive to have a skilled technician spend time diagnosing and fixing a specific problem rather than a less skilled individual who can swap a single component.


Its not really a weird perspective. Where are most of Apple's Mac sales? In wealthy areas like the US and EU. The hourly rate for someone able to diagnose a problem and do the required soldering is pretty high. Those customers don't want to wait a couple days, especially with no guarantee that the diagnosis step will be successful. It makes more sense for them to replace an entire board or even the entire Mac, which at worst would take a day - to allow for overnight shipment if they don't have the required part/model on hand. The you ship what's broken somewhere with much lower cost engineers like China or India where it can be diagnosed, repaired and recertified to be sold as reconditioned.

This is how it is if your phone breaks other than a screen or battery, and no one thinks there is anything wrong with that. The only reason they think there is anything wrong with this model for laptops or desktops is because they are used to a different way. People used to be used to doing their own auto repairs, and when I was a kid it wasn't unusual on a weekend to see a neighbor with his car jacked up changing the oil or doing a tuneup or whatever. I can't remember the last time I saw someone working on their car at home. If they are they are constantly washing it or they are customizing it in some way - i.e. the automotive equivalent of the hardcore gamer / enthusiast crowd from the PC world.
 
  • Like
Reactions: Mopetar

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
It is my personal opinion that Apple has made the right call on their lower end Macs with respect to their internal SSD storage. If you need more storage, get an external TB4 SSD. Good ones can sustain 2500MB/sec throughput which is good enough for most personal usage. Yes, it's not as fast as a top-end PCIe 4.0 4 lane NVME drive, but the functional difference in the vast majority of instances is barely measurable. I think Apple's bigger sin is in how small the RAM allotments are for their lower end devices, but it really doesn't seem to negatively affect most casual users.

I know that we like to open up our computers and tinker with them, but, that's not my typical experience with Mac users. The home and business users just want an inoffensive box that "just works" and can be replaced when it's old or breaks. The pro and prosumer users know what they are getting and can plan accordingly. I may not be a fan of their lower end devices being such a walled garden both in software and hardware, but then, I'm not a mac user either, so my opinion doesn't much matter. For their business model and audience, they are doing what makes the most sense.
 

kschendel

Senior member
Aug 1, 2018
261
191
116
It's a bit of a weird perspective, but at a certain point even expensive equipment can become less expensive to replace outright than to do repair work on. Labor costs are what ultimately drive these decisions.

Not a weird perspective at all. That's been the trajectory of almost every significant consumer product class that I can think of. In the 60's I could take a stack of tubes to the drugstore tube tester and have a reasonable chance of successfully fixing the TV. Not too many years later, TV repair consisted largely of board swapping. Similarly with cars, as someone else mentioned. Phones, radios (which barely exist as standalone items any more), oven/range displays, washer/dryer control boards or assemblies, etc etc.
 

GC2:CS

Junior Member
Jul 6, 2018
23
19
81
Looks like there is a special version of M2 Max.

The 38core GPU version in 16 Mac Book pro reaches 3,7 Ghz on P cores instead of 3,5 in all other setups.
This produces 2000+ single core geek bench results.


It is interesting too see differentiation come this silently, but it is inevitable that Apple will eventually tune the chip for each device. 3.46 Ghz in an iPhone and 3,67 in a Mac just does not seem right.

There is also a "replacement" for anadtech review of A16 on this youtube chanel.


More or less confirms that A16 is an incremental update in CPU and GPU, unlike previous chips which looked like small upgrades from Apple marketing, but turned out to be changed profoundly.

Does that mean we might get M3 generation based right of A17 ?
M2 gen and A16 gen is almost equal in all metrics.
 
  • Like
Reactions: Tlh97

Doug S

Platinum Member
Feb 8, 2020
2,205
3,409
136
Does that mean we might get M3 generation based right of A17 ?
M2 gen and A16 gen is almost equal in all metrics.


M1 used A14 cores, and the same process as A14. M2 used A15 cores and the same process as A15. I would expect M3 to use A17 cores since both will use some sort of N3 variant (some sort of Apple customized version of N3 that incorporates elements of N3E but is neither N3 or N3E if I had to guess)

Supposedly Apple had some big updates for the GPU planned for A16 but they didn't get it working in time, and had to ship it with A15's GPU. So A17/M3 should see big GPU improvements. We are well overdue for CPU improvements, since there has not been a whole lot of changes between A14 and A16, other than under the hood stuff needed for bigger Macs like additional bits of physical and virtual address, that sort of thing.
 
Jul 27, 2020
15,776
9,833
106
People worry way too much about hitting the wear limits on their NAND. I put an SSD in my Tivo and despite 24x7 recording on four tuners (I don't ever put it in standby mode) for two years SMART tells me I have only used 20% of its capacity.
Which SSD model? Tuners would be doing mostly sequential writes. What destroys NAND is 4K random writes coz of write amplification.
 

mikegg

Golden Member
Jan 30, 2010
1,740
406
136
More evidence that Apple plans yearly Apple Silicon updates:

"The M2 family was really now about maintaining that leadership position by pushing, again, to the limits of technology. We don't leave things on the table," says Millet. "We don't take a 20% bump and figure out how to spread it over three years...figure out how to eke out incremental gains. We take it all in one year; we just hit it really hard. That's not what happens in the rest of the industry or historically."


Other evidence/speculation include:
  • Rumors were that M2 SoC was ready much earlier but was held back by the late new Air design.
  • Reports and video file name suggest that the M2 Pro/Max were ready as early as October 2022, which would have made it one year since M1 Pro/Max
  • M2 used A15 cores, which suggests Apple wants to take advantage of the iPhone SoC design updates each year
  • Reports are that M2 will be a short-lived generation because M3 is expected to launch on time
  • Many speculators point to iPad SoCs, which were not updated yearly. However, the base M chips go into far more devices than old iPad SoCs. This makes it more economical to design and launch new SoCs yearly.
  • M1 launched in the same quarter as A14, which suggests Apple develops the base M and the A series in tandem
  • Apple has been fighting supply chain issues in the last 3 years, which caused many delays in new designs shipping.
Why is this important? Because gaining 10-20% in performance once every two years is not impressive. But once a year? Now that's impressive and exciting.
 
Last edited:
  • Haha
Reactions: scineram

soresu

Platinum Member
Dec 19, 2014
2,617
1,812
136
More evidence that Apple plans yearly Apple Silicon updates:




Other evidence/speculation include:
  • Rumors were that M2 SoC was ready much earlier but was held back by the late new Air design.
  • Reports and video file name suggest that the M2 Pro/Max were ready as early as October 2022, which would have made it one year since M1 Pro/Max
  • M2 used A16 cores, which suggests Apple wants to take advantage of the iPhone SoC design updates each year
  • Reports are that M2 will be a short-lived generation because M3 is expected to launch on time
  • Many speculators point to iPad SoCs, which were not updated yearly. However, the base M chips go into far more devices than old iPad SoCs. This makes it more economical to design and launch new SoCs yearly.
  • M1 launched in the same quarter as A15, which suggests Apple develops the base M and the A series in tandem
  • Apple has been fighting supply chain issues in the last 3 years, which caused many delays in new designs shipping.
Why is this important? Because gaining 10-20% in performance once every two years is not impressive. But once a year? Now that's impressive and exciting.
It's not really a huge jump when they've been doing the same thing in the mobile market with Ax SoCs for over a decade.

It stands to reason that they will want to maximise R&D recoup for each CPU and GPU µArch iteration by spreading them across as many market segments as possible.
 
  • Like
Reactions: Mopetar

Mopetar

Diamond Member
Jan 31, 2011
7,797
5,899
136
From what I can find M2 is ARMv8.5A whereas the A15 uses the slightly older ARMv8.4A ISA.

I don't know how much practical difference there is between the two, but supposed the M2 has for support for any instructions added in the 8.5 update. The ARM blog page on what v8.5 added doesn't make it seem like a major overhaul. Some instructions for better random numbers and a few things to make it harder for the chip to be exploited by certain types of attacks, but nothing earth shattering.