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

 

[defferoo]

But reviewers do.

and you think Apple cares enough to change their SoC roadmap cause of some reviewers? they obviously compare against all of their competitors internally, but that's not going to change their release cadence.

@Eug the reason why Apple hasn’t launched the M4 in Macs yet is because of optics. They now want to release all the MBP’s at the same time that means the Max and Pro have to be ready.

That won’t happen till November, bigger chips require more time for validation etc.

Yeah, for Mac lineup, they probably don't want their cheapest laptop to have faster CPU performance than their mid-level Pro laptop. I think a fully enabled M4 beats the 11 core M3 Pro in GB MT. They need the M4 Pro/Max chips to be ready before they can release M4 on Mac.

 

[johnsonwax]

Apple Mac shipments worldwide by quarter 2023 | Statista

In the fourth quarter of 2023, Apple Mac shipments stood at 6.3 million units worldwide, an increase from the 5.9 million Mac shipments that were recorded in the fourth quarter of the previous year.

www.statista.com

The demand peaks are fine but it's the sales slump nosedives that might cause inventory to build up. Obviously I don't have any real data but if I were an Apple investor, I would be interested in Apple's official strategy regarding unsold inventory.

Been lurking a while. Created an account to answer this question. 27 year AAPL long investor here.

Your understanding of Apple operations is very speculative and in the wrong direction. To start, Apple doesn't have warehouses, effectively. Product is shipped from the factory directly to consumers or directly to Apple Stores, which serve as local warehouses. They are entirely air freight so there is minimal latency from order placement to arrival eliminating the need to stockpile to smooth that out as you would have with a 30 day container shipment model. Tim Cook quote on the issue:

Inventory is fundamentally evil. You kind of want to manage it like you're in the dairy business. If it gets past its freshness date, you have a problem.

If you want to know why Apple goes for minimalist product size and packaging, how many you can fit in a 747 and in the back of an Apple Store is a driving consideration. The Mac Mini, way back in 2005 was designed this way to reduce overall costs. It as cheaper for them to switch to expensive 2.5" drives than to cover the shipping and inventory costs of a larger enclosure to accommodate a cheaper 3.5" drive. This is just how they roll.

So in terms of finished devices, they are effectively 100% BTO. Apple has for the last 20 years never exceeded 10 days of inventory to my knowledge, and that reflects the time in the belly of a plane and in the back of an Apple Store. They have a bit of warehouse operations for spare part inventory, which is attached to their repair centers. So, essentially any time Apple makes a product announcement they usually don't hold enough inventory to coast from announcement to availability of the new product and sometimes hit a small product unavailability window of a few days if they are immediately discontinuing the old product. Put another way, it's a small enough quantity that they can just eat the inventory, use it for warranty replacement, for refurb (I've seen people order refurb and get new), or use it for the new interns device. It's a really tiny number.

That doesn't address components though. Let's say there is a big falloff in M3 demand, which is already in production at TSMC, Apple will simply roll those chips into the next iPad Air which will likely ship with the M3 anyway and likely still be in production after the M4 ships (current iPad Air has an M2 in it). Apple cascades components down their device family pretty regularly that way. A big miss on A processors might trigger an AppleTV update. They can manage demand for those pretty easily through incentives. Apple Back to School was a pretty good tool for moving lagging iPod demand by bundling it with a new Mac. If iPod wasn't lagging, then it'd be free Apple Care. Their full product spectrum is frequently leveraged to address these kinds of problems (people have no understanding for why Apple bought Beats). It's also why Apple carries so few SKUs. It's a lot harder to manage this stuff across 30 different models of phones that all have different components than it is across 3 models of phones which mostly share components, plus another two which are just last years models with the price cut. Apple plays tall.

In addition to these tools, Apple has at least one other. We tend to be cognizant of the product matrix in the US/EU, but Apple deviates from this in other markets. It's not uncommon for Apple to keep an old iPhone for sale in a price-conscious market like India for an additional year or more. For how long isn't really a matter of product announcements but running out a production contract. They'll make x million additional iPhone 14s at pretty low cost because the components are all EOL, and offer them for sale until they're gone. In the US they have historically done this in the education market as well - keep the M3 iMac production going until they exhaust the M3 component agreement and make them available to education for $100 off or something. Makes a big difference if you're building out a general use lab. These were always hit or miss if there would be such a model, but there often is. Sometimes they're available to K-12 and higher ed, sometimes just K-12 - depends on how much of that component contract they need to cover.

Inventory management was a big topic of conversation back in the 90s, especially around Apple as pre-Cook Apple absolutely sucked at it (like B-school case study bad at it), but pretty quickly things changed and suddenly beleaguered Apple was beating Dell, the then king of just-in-time production, and basically rewrote the book on how to do it. The retail stores (which weren't supposed to succeed) became a key component, the air freight (which was supposed to be too expensive) became a key component. It influenced product design and packaging. It influenced moving away from physical media to free up space in the stores for the expanding Apple hardware lineup. It influenced how these contracts were written as Apple got larger and how much flexibility Apple had in the number of units, and so on. So, for instance, Apple wants to buy components from you, but you don't have the capacity to meet Apple's volume. Apple, having more cash than JP Morgan will give you the cash up front to build a new factory and you will repay that loan in product. Apple will provide a purchase guarantee of $x or y units, demand first refusal for anything that your factory produces (so that they get the best camera sensor instead of Samsung) for maybe 3 years, and in return they will get unprecedented flexibility in terms of orders after that purchase guarantee. You in turn get a 0% loan for a new factory and a guarantee you have a buyer for everything that factory produces until that loan is repaid. Apple gets supply chain control. It's not a bad deal.

Apple is still the absolute best in the world at this stuff. There are some things Apple investors should be concerned about, supply chain management ain't one of them.

 

[johnsonwax]

No chance. if they increase specs I would expect a $200 price increase at minimum.

Since I have a shiny new account to break in, I'll address this as well.

Apple doesn't increase prices. I know that sounds incredible, but really, they pretty much don't. Apple sets price points and designs to price point. The design team gets a set of requirements and they have latitude in terms of how to hit the price point. Apple may well specify 16GB as a minimum due to the increased compute needs of these learning models, so it's up to the engineers to figure out how to build that and stay at the stated price point (not that it'd be hard, Apples cost on 8GB RAM is what, $8 tops?). Sometimes they'll shift it slightly up or down depending on if there's a particularly good or poor value for money - for a $50 increase we can switch from a higher capacity HDD to a lower capacity SSD and get an order of magnitude improvement on read/write speeds, etc. And then they'll revert the price the following year when SSD prices come down.

Here's my evidence. Go look at the price for the base model iMac:

iMac​

From $1299 or $108.25/mo. for 12 mo.**
Apple M3 chip
8‑core CPU
8‑ or 10‑core GPU
8GB to 24GB unified memory
256GB to 2TB storage
4.5K Retina display

Go look up what the base price for the iMac was when it was first introduced in 1998:

G3/233 processor
32MB of standard memory
4GB of disk storage
24x CD-ROM drive
100Mb Ethernet port
33.6Kb modem
15-inch display that was 1024x768
$1299

The iMac hasn't moved more than $100 in price over 26 years. Apple sets a price and then value-adds to justify the price. If they go to 16GB RAM, it'll be $1299. They rely on inflation to make the product more affordable over time, and the even faster rate of technology cost decline to be able to consistently value-add.

Don't confuse how Apple upsells their products with how they set their product strategy. They are doing completely different things. Apple has the Hershey bar and Arizona Ice Tea beat for price stability. That's a quarter century of price stability. I know they have a reputation of always increasing prices - they very rarely do. They will increase the ceiling on what you can upsell to, but the base rarely moves. iPhone has moved up more than anything else. Others have moved down. MacBook Air from $1799 in 2008 to $1099 now.

 

[johnsonwax]

Regarding the timing of the M4 arrival and how their product strategy may evolve, I'm reminded of this article:

Brett Simpson, senior analyst at Arete Research, claimed that TSMC and Apple have reached a special deal, through which Apple will only pay for functional circuits, rather than paying standard pricing for the entire wafer. If the 3nm yield rate improves to a regular 70%, the deal may be adjusted so that Apple pays up to $17,000 per wafer in the second half of 2024.

My interpretation here was that TSMC was about to violate their contract with Apple regarding yield, and stepped up to eat Apple's yield (talk about leverage, that's cost that's being indirectly passed back to Qualcomm, AMD, Nvidia). These problems didn't improve enough over the M3 product family so Apple pushed to roll M4 out as quickly as they could, starting with their highest unit volume device - the iPad Pro, which got off of that node even before the second half of 2024 referenced in the article. 'A' series is stuck on its own clock so that's just going to have to play out normally, but Apple can cut and run on the M3 as quickly as TSMC can get N3E volume to meet Apple's needs. My guess is that'll all be reserved for A18, which pushes M4 Pro, etc. out to the rumored November, with the possibility that they are going to move to N3P for the higher end product line which probably pushes that out a bit further - it seems odd that we're talking this timetable and not taking advantage of N3P for the Pro Macs, if we are to assume that M5 will wait for 2nm. It doesn't make sense that Apple would skip that opportunity.

Yes, having M4 Macs outperforming M3 Pro is a little bit of a problem but not a new one. The MBP will still outperform on GPU by a wide margin, has higher RAM options, etc. There's still a market for it. Besides if having iPad faster than the MBP isn't a problem, having a MBA faster than a MBP shouldn't be either. I could see M4s Macs rolling out as there's capacity for them. They've been here before. Remember when the MacBook Air was legit like 2x faster than the Intel Mac Pro rather than just 8% faster? They'll manage.

Remember, Apple base iPhone is still A16. So this first 3nm process is only covering M3/Pro/Max and iPhone Pro. None of the iPads either. They went from M2 to M4 in iPad Pro and iPad/mini never went to A17. My guess is that this years iPhone won't adopt A17 and stay a generation behind iPhone Pro like last year, that Apple will have A18 across the board. Apple is getting everything on N3E and later as quickly as TSMC has capacity. There's no coasting on A17/M3 across products because it's a turd in terms of yield. Apple pulling the emergency chute here. If that pushes other M4 products out to make room for A production, so be it.

So I retract my previous statement that Apple could slide M3 into a trailing product. I think they're out. Kill M3 production as soon as practical and don't look back. Not the only explanation for their product strategy here, but the most straightforward one.

 

[Eug]

Since I have a shiny new account to break in, I'll address this as well.

Apple doesn't increase prices. I know that sounds incredible, but really, they pretty much don't. Apple sets price points and designs to price point. The design team gets a set of requirements and they have latitude in terms of how to hit the price point. Apple may well specify 16GB as a minimum due to the increased compute needs of these learning models, so it's up to the engineers to figure out how to build that and stay at the stated price point (not that it'd be hard, Apples cost on 8GB RAM is what, $8 tops?). Sometimes they'll shift it slightly up or down depending on if there's a particularly good or poor value for money - for a $50 increase we can switch from a higher capacity HDD to a lower capacity SSD and get an order of magnitude improvement on read/write speeds, etc. And then they'll revert the price the following year when SSD prices come down.

Here's my evidence. Go look at the price for the base model iMac:



Go look up what the base price for the iMac was when it was first introduced in 1998:



The iMac hasn't moved more than $100 in price over 26 years. Apple sets a price and then value-adds to justify the price. If they go to 16GB RAM, it'll be $1299. They rely on inflation to make the product more affordable over time, and the even faster rate of technology cost decline to be able to consistently value-add.

Don't confuse how Apple upsells their products with how they set their product strategy. They are doing completely different things. Apple has the Hershey bar and Arizona Ice Tea beat for price stability. That's a quarter century of price stability. I know they have a reputation of always increasing prices - they very rarely do. They will increase the ceiling on what you can upsell to, but the base rarely moves. iPhone has moved up more than anything else. Others have moved down. MacBook Air from $1799 in 2008 to $1099 now.

For the iPad Pro, they took the iPhone approach and raised prices significantly. I was not surprised though, but that price increase was done in such a way that it was palatable to some of us: Form factor update with landscape Face ID, new Magic Keyboard and Pencil Pro, dual-stack OLED, and doubled base storage, etc.

 

[Doug S]

Regarding the timing of the M4 arrival and how their product strategy may evolve, I'm reminded of this article:


My interpretation here was that TSMC was about to violate their contract with Apple regarding yield, and stepped up to eat Apple's yield (talk about leverage, that's cost that's being indirectly passed back to Qualcomm, AMD, Nvidia). These problems didn't improve enough over the M3 product family so Apple pushed to roll M4 out as quickly as they could, starting with their highest unit volume device - the iPad Pro, which got off of that node even before the second half of 2024 referenced in the article. 'A' series is stuck on its own clock so that's just going to have to play out normally, but Apple can cut and run on the M3 as quickly as TSMC can get N3E volume to meet Apple's needs. My guess is that'll all be reserved for A18, which pushes M4 Pro, etc. out to the rumored November, with the possibility that they are going to move to N3P for the higher end product line which probably pushes that out a bit further - it seems odd that we're talking this timetable and not taking advantage of N3P for the Pro Macs, if we are to assume that M5 will wait for 2nm. It doesn't make sense that Apple would skip that opportunity.

Yes, having M4 Macs outperforming M3 Pro is a little bit of a problem but not a new one. The MBP will still outperform on GPU by a wide margin, has higher RAM options, etc. There's still a market for it. Besides if having iPad faster than the MBP isn't a problem, having a MBA faster than a MBP shouldn't be either. I could see M4s Macs rolling out as there's capacity for them. They've been here before. Remember when the MacBook Air was legit like 2x faster than the Intel Mac Pro rather than just 8% faster? They'll manage.

Remember, Apple base iPhone is still A16. So this first 3nm process is only covering M3/Pro/Max and iPhone Pro. None of the iPads either. They went from M2 to M4 in iPad Pro and iPad/mini never went to A17. My guess is that this years iPhone won't adopt A17 and stay a generation behind iPhone Pro like last year, that Apple will have A18 across the board. Apple is getting everything on N3E and later as quickly as TSMC has capacity. There's no coasting on A17/M3 across products because it's a turd in terms of yield. Apple pulling the emergency chute here. If that pushes other M4 products out to make room for A production, so be it.

So I retract my previous statement that Apple could slide M3 into a trailing product. I think they're out. Kill M3 production as soon as practical and don't look back. Not the only explanation for their product strategy here, but the most straightforward one.

Yes that's another consideration I'd forgot. It was very widely reported that TSMC was charging Apple for "known good dies" for A17P/M3 made on N3B rather than the typical per wafer pricing, but there were rumors that the deal was time/volume limited in some fashion so that Apple would be "encouraged" to move to N3E when that node was available.

Moving to M4 on iPad Pro is only a partial solution, as they say here Apple is still shipping N3B in the form of M3 for Macs and A17P for iPhone. It has been very widely reported Apple will have A18 and A18P both made on N3E for iPhone 16, and since Apple only keeps the base iPhone around as "last year's model" which would be iPhone 15 which isn't N3B they'll be done wity N3B for iPhones in exactly three months. As he points out they went from M2 to M4 for iPad Pro so there's no M3 for "last year's model". None Pro iPads aren't on N3B, so that's iPhone and iPad taken care of as of late September.

So what to do with the Macs? If they REALLY want to get off N3B entirely then on models that are currently M3 and have M2 as "last year's model" they'd upgrade them to M4 but continue selling M2 as "last year's model" just with a bigger discount to make up for it being two years old. If they actually do that then we'll know just how hard TSMC encouraged/pushed them off N3B! That would leave only MBA which got M3 only a few months ago, but they go that far on other Macs then I would anticipate the M3 MBA is replaced severals months short of the one year mark.

I disagree with the article about taking advantage of N3P for M4 Pro/Max. If N3P was starting mass production in July maybe that would be possible but my understanding is that it is following N3E by a year so they'd be well into 2025 before any N3P chips can ship in product. If the goal is to get off N3B you don't wait around for a node that isn't delivering finished chips until probably Q2 2025.

So while this is all interesting, it still doesn't offer any new explanations for why Apple isn't upgrading Macs to M4 now. I get that they have to build up stock of A18/A18Ps for the iPhone launch but that's three months away, there should be plenty of capacity to make all the M4s Apple needs. I still think the only explanation is that M4 Pro and Max aren't ready, and they can't offer just the entry level M4 in all those other Macs that offer higher end options.

So why aren't they ready? One thing caught my eye today - the claim that A18P will have a better NPU than M4. They would have had a deadline for M4 tapeout around this time last year, at the very latest, to make their planned iPad Pro launch. The A18 tapeout deadline would have been 4-5 months later. I think with the AI hype building after ChatGPT launched in late 2022, and increasing when ChatGPT 4 came out a few months later Apple may have wanted to add a better NPU in A18P, and perhaps they decided to add that better NPU in the higher end versions of M4 so the tapeout of M4P/M4M was pushed back to make that change.

 

[defferoo]

Regarding the timing of the M4 arrival and how their product strategy may evolve, I'm reminded of this article:


My interpretation here was that TSMC was about to violate their contract with Apple regarding yield, and stepped up to eat Apple's yield (talk about leverage, that's cost that's being indirectly passed back to Qualcomm, AMD, Nvidia). These problems didn't improve enough over the M3 product family so Apple pushed to roll M4 out as quickly as they could, starting with their highest unit volume device - the iPad Pro, which got off of that node even before the second half of 2024 referenced in the article. 'A' series is stuck on its own clock so that's just going to have to play out normally, but Apple can cut and run on the M3 as quickly as TSMC can get N3E volume to meet Apple's needs. My guess is that'll all be reserved for A18, which pushes M4 Pro, etc. out to the rumored November, with the possibility that they are going to move to N3P for the higher end product line which probably pushes that out a bit further - it seems odd that we're talking this timetable and not taking advantage of N3P for the Pro Macs, if we are to assume that M5 will wait for 2nm. It doesn't make sense that Apple would skip that opportunity.

Yes, having M4 Macs outperforming M3 Pro is a little bit of a problem but not a new one. The MBP will still outperform on GPU by a wide margin, has higher RAM options, etc. There's still a market for it. Besides if having iPad faster than the MBP isn't a problem, having a MBA faster than a MBP shouldn't be either. I could see M4s Macs rolling out as there's capacity for them. They've been here before. Remember when the MacBook Air was legit like 2x faster than the Intel Mac Pro rather than just 8% faster? They'll manage.

Remember, Apple base iPhone is still A16. So this first 3nm process is only covering M3/Pro/Max and iPhone Pro. None of the iPads either. They went from M2 to M4 in iPad Pro and iPad/mini never went to A17. My guess is that this years iPhone won't adopt A17 and stay a generation behind iPhone Pro like last year, that Apple will have A18 across the board. Apple is getting everything on N3E and later as quickly as TSMC has capacity. There's no coasting on A17/M3 across products because it's a turd in terms of yield. Apple pulling the emergency chute here. If that pushes other M4 products out to make room for A production, so be it.

So I retract my previous statement that Apple could slide M3 into a trailing product. I think they're out. Kill M3 production as soon as practical and don't look back. Not the only explanation for their product strategy here, but the most straightforward one.

still think the M4 Pro/Max aren’t ready which is why they haven’t transitioned Macs to M4. they haven’t had a Macbook Air faster than the Macbook Pro since they transitioned to Apple Silicon. People aren’t likely to be cross-shopping an iPad Pro and Macbook Pro so it matters less for the iPad, but that’s not true of the Macbook Air and lower end macbook Pros.

 

[johnsonwax]

Yes that's another consideration I'd forgot. It was very widely reported that TSMC was charging Apple for "known good dies" for A17P/M3 made on N3B rather than the typical per wafer pricing, but there were rumors that the deal was time/volume limited in some fashion so that Apple would be "encouraged" to move to N3E when that node was available.

Moving to M4 on iPad Pro is only a partial solution, as they say here Apple is still shipping N3B in the form of M3 for Macs and A17P for iPhone. It has been very widely reported Apple will have A18 and A18P both made on N3E for iPhone 16, and since Apple only keeps the base iPhone around as "last year's model" which would be iPhone 15 which isn't N3B they'll be done wity N3B for iPhones in exactly three months. As he points out they went from M2 to M4 for iPad Pro so there's no M3 for "last year's model". None Pro iPads aren't on N3B, so that's iPhone and iPad taken care of as of late September.

So what to do with the Macs? If they REALLY want to get off N3B entirely then on models that are currently M3 and have M2 as "last year's model" they'd upgrade them to M4 but continue selling M2 as "last year's model" just with a bigger discount to make up for it being two years old. If they actually do that then we'll know just how hard TSMC encouraged/pushed them off N3B! That would leave only MBA which got M3 only a few months ago, but they go that far on other Macs then I would anticipate the M3 MBA is replaced severals months short of the one year mark.

I think they'd retire last years model. It's unclear how much holding M2 products around is a function of avoiding M3/A17 yield problems and how much is desirable. Apple isn't perfectly consistent on this - they could stick with an M2/M4 split, they could just drop the M2.

I mean, Apple made all kinds of trade-offs to get through the Apple Silicon transition. If N3B really is that kind of a problem, they'll simply have to do that again here - the product strategy will simply have to adapt. And it's still pretty hard to figure out what the nominal Apple Silicon product strategy is going to be - we had the costs of introducing the various M1 flavors, M2 was pretty normal and Ultra arrived, M3 skipped Ultra (intended?) and M4 showed up too soon. Not enough to establish a pattern. But I think skipping from M2 to M4 is indicative that M3 is a serious problem, if only a cost/production volume one.

I disagree with the article about taking advantage of N3P for M4 Pro/Max. If N3P was starting mass production in July maybe that would be possible but my understanding is that it is following N3E by a year so they'd be well into 2025 before any N3P chips can ship in product. If the goal is to get off N3B you don't wait around for a node that isn't delivering finished chips until probably Q2 2025.

My understanding was that N3P was volume production in 2nd half 2024.

So while this is all interesting, it still doesn't offer any new explanations for why Apple isn't upgrading Macs to M4 now. I get that they have to build up stock of A18/A18Ps for the iPhone launch but that's three months away, there should be plenty of capacity to make all the M4s Apple needs. I still think the only explanation is that M4 Pro and Max aren't ready, and they can't offer just the entry level M4 in all those other Macs that offer higher end options.

The last few versions have split the iPhone/iPhone Pro by processor generations. That would imply that iPhone Pro would go A18 and iPhone would go A17. If we go with the hypothesis that Apple wants off of N3B entirely, then they are going to go A18 across both iPhone and Pro. That's an increase of 100 million units for N3E over N3B, front loaded to launch. Does TSMC have *that* kind of capacity ready to go? They can always bin the the iPhone units and drop a core, but that's still a big unit increase on the new equipment. They can't stick with the A16 due to Apple Intelligence - requires an A17 or better.

So why aren't they ready? One thing caught my eye today - the claim that A18P will have a better NPU than M4. They would have had a deadline for M4 tapeout around this time last year, at the very latest, to make their planned iPad Pro launch. The A18 tapeout deadline would have been 4-5 months later. I think with the AI hype building after ChatGPT launched in late 2022, and increasing when ChatGPT 4 came out a few months later Apple may have wanted to add a better NPU in A18P, and perhaps they decided to add that better NPU in the higher end versions of M4 so the tapeout of M4P/M4M was pushed back to make that change.

Yeah, my guess is that M4 was ready for production before A18, TSMC had capacity, Apple wanted to use that so started with M4, pushed the iPad so the Mac wouldn't get horribly stretched out and split for a long time between base M4 and M3 Pro/Max while they shifted to A18 once it came ready, and then 6-8 months later got around to M4 Pro/Max. So we get a kick-ass iPad, shift to A18 - more than originally planned due to iPhone jumping from A16-A18 - and the Macs just have to wait resulting in a somewhat more compressed M4/Pro/Max rollout.

I'm not sure the NPU plays out as you suggest. The beefier NPU in the A series was generally to accommodate computational photography needs, which the M series didn't need. As such the A17 NPU is in the ballpark of the M4 already. If the M4 is sufficient for Apple Intelligence, then the A series NPU doesn't need to increase _for that purpose_. It might need to increase for computational photography needs - either due to larger sensor, new features, etc. but Apple had that information at the usual time. But for the purposes of Apple Intelligence, the M4 and A17 should be sufficient. Apple might want more NPU cores in the Pro/Max products simply because those are more likely to be leaning on more of the generative stuff in Logic, and presumably in an updated FCP that they might want to avoid having to ship onto servers if possible (lemme upload my ProRes RAW to your server to remove that boom mike sounds aboslutely miserable), but I don't see why A18 would need a last minute NPU push because of AI, nor why Pro/Max wouldn't know their needs at the same time that M4 did. The M4 design needed to predate the Pro/Max no matter what, and coming in ahead of A18 could just be a matter of having things for the team to work on concurrently. Also might have been a pants on fire response to the aforementioned yield crisis, knowing getting A18 done sooner wouldn't yield anything due to iPhone being on a strict calendar but the M products being flexible. Again, we don't have enough of a track record to draw inferences from. All we know that is that these timelines are long, and they can't be compressed too much. But Apple can make decisions like cancelling the A17 going into iPhone and going with A18 on both iPhone and Pro and asking TSMC if they can accommodate an additional mountain of N3E wafers in 2024, and pushing Macs (the lowest volume product) back if necessary. You don't need any design theories to explain that one - that's just production capacity and how fast they can ramp. It cements the iPhone, the most important product, it gets some M4 production going when there's capacity but no A18 design ready and it pushes the rest of M4 well back because they're going to make 60% more A18s than anticipated - a decision they could have made a year ago when the problems with yield first surfaced.

 

[johnsonwax]

still think the M4 Pro/Max aren’t ready which is why they haven’t transitioned Macs to M4. they haven’t had a Macbook Air faster than the Macbook Pro since they transitioned to Apple Silicon. People aren’t likely to be cross-shopping an iPad Pro and Macbook Pro so it matters less for the iPad, but that’s not true of the Macbook Air and lower end macbook Pros.

I still don't think a mismatch across the product line is that big of a problem for Apple. People aren't buying Pros for better single core performance, because they don't have better single core. They do have more cores and still will have more cores. They have much better GPUs. They have more RAM capacity, ports, how many monitors they can drive. The place where the M4 is faster than the M3 Pro is probably the one place nobody is choosing a MBP over an Air. It's a little embarrassing, but nothing more than that. Pro users are generally pretty savvy about what they need and whether it's worth buying or waiting. It's the non-pro users that are usually walking into the store to buy an Air with no clue where they are in the product cycle and will buy an Air the day before the new ones are announced. Pro users usually don't do that, and if they do, it's because they can't wait.

I expect that Pro/Max aren't ready, and if N3P is volume production 2nd half of this year, I'd expect they are designed for that process, so we'll get a split between N3E for M4 and N3P for the bigger stuff (they did this before). Lots of benefits to doing that. But if Apple was planning that, they were almost certainly going to roll out M4 Macs before the Pro/Max ones simply due to the processes hitting production at different times. Apple surely knew this 2+ years ago. The roadmap doesn't always hit on time, but it usually hits in sequence.

 

[igor_kavinski]

@Eug , would you be willing to run this sheet on your M4 device? https://www.overclock.net/threads/excel-benchmark.1642077/page-10?nested_view=1&sortby=oldest

Run it with this: https://apps.apple.com/us/app/collabora-office/id1440482071 because it's supposed to be faster than Excel.

Macros will need to be ENABLED (sorry, typo when I wrote disabled before).

 

[richardskrad]

The M4 is the highest single core CPU in the world and it’s in a ridiculously thin tablet without cooling. I don’t understand where this idea that Apple has lost most of their chip design talent and is in trouble rhetoric is coming from. Apple still makes the world’s most advanced and efficient consumer chips, years after the M1.

 

[Eug]

@Eug , would you be willing to run this sheet on your M4 device? https://www.overclock.net/threads/excel-benchmark.1642077/page-10?nested_view=1&sortby=oldest

Run it with this: https://apps.apple.com/us/app/collabora-office/id1440482071 because it's supposed to be faster than Excel.

Macros will need to be ENABLED (sorry, typo when I wrote disabled before).

Hmmm… I’m not keen on running a third party Excel file with macros on.

 

[defferoo]

I still don't think a mismatch across the product line is that big of a problem for Apple. People aren't buying Pros for better single core performance, because they don't have better single core. They do have more cores and still will have more cores. They have much better GPUs. They have more RAM capacity, ports, how many monitors they can drive. The place where the M4 is faster than the M3 Pro is probably the one place nobody is choosing a MBP over an Air. It's a little embarrassing, but nothing more than that. Pro users are generally pretty savvy about what they need and whether it's worth buying or waiting. It's the non-pro users that are usually walking into the store to buy an Air with no clue where they are in the product cycle and will buy an Air the day before the new ones are announced. Pro users usually don't do that, and if they do, it's because they can't wait.

I expect that Pro/Max aren't ready, and if N3P is volume production 2nd half of this year, I'd expect they are designed for that process, so we'll get a split between N3E for M4 and N3P for the bigger stuff (they did this before). Lots of benefits to doing that. But if Apple was planning that, they were almost certainly going to roll out M4 Macs before the Pro/Max ones simply due to the processes hitting production at different times. Apple surely knew this 2+ years ago. The roadmap doesn't always hit on time, but it usually hits in sequence.

Apple is ALL about segmenting their products in a way that it entices people to upgrade to the next tier. a MacBook Air with higher single-core and multi-core CPU performance than their M3 Pro MacBook Pros is a situation they want to avoid if they can. GPU performance matters a lot less than CPU to most people. There's literally nothing pushing them to release M4 Macs except maybe TSMC trying to get them off N3B so it makes more sense for them to wait until the full lineup is ready.

 

[FlameTail]

How did Apple put an SoC that consumes upto 30W each for CPU and GPU, in a FANLESS IPAD!?


@ikjadoon

 

[coercitiv]

How did Apple put an SoC that consumes upto 30W each for CPU and GPU, in a FANLESS IPAD!?

Through the wonder of temperature target.

 

[Mopetar]

If you don't give it 30W of power it's not that difficult really.

You could put a 14900K (which can guzzle over 300W) into a device designed around 30W, but it would mostly be a waste and no better than some chip with an intended TDP closer to that amount.

CPUs only dissipate heat in direct proportion to the power they draw. Feed a CPU 10W of power and it only needs a cooling solution that can handle those 10 watts.

A race car can handle driving on a 25 MPH street just fine even though it may been designed to go over 200 MPH on a speedway.

 

[igor_kavinski]

A race car can handle driving on a 25 MPH street just fine even though it may been designed to go over 200 MPH on a speedway.

Not sure about that analogy. Wouldn't the rapid acceleration of the race car be problematic? Or do street legal race cars have separate normal and racing modes?

 

[defferoo]

How did Apple put an SoC that consumes upto 30W each for CPU and GPU, in a FANLESS IPAD!?

throttling my friend, throttling.

Not sure about that analogy. Wouldn't the rapid acceleration of the race car be problematic? Or do street legal race cars have separate normal and racing modes?

you could just limit how hard you press on the pedal? not sure I understand the issue

 

[name99]

Since I have a shiny new account to break in, I'll address this as well.

Apple doesn't increase prices. I know that sounds incredible, but really, they pretty much don't. Apple sets price points and designs to price point. The design team gets a set of requirements and they have latitude in terms of how to hit the price point. Apple may well specify 16GB as a minimum due to the increased compute needs of these learning models, so it's up to the engineers to figure out how to build that and stay at the stated price point (not that it'd be hard, Apples cost on 8GB RAM is what, $8 tops?). Sometimes they'll shift it slightly up or down depending on if there's a particularly good or poor value for money - for a $50 increase we can switch from a higher capacity HDD to a lower capacity SSD and get an order of magnitude improvement on read/write speeds, etc. And then they'll revert the price the following year when SSD prices come down.

To add to this point (and great recent posts BTW) I remind you of the infamous Asymco iPhone pricing chart:

The Poetry of Pricing

iPhone 15 has just been released and, as usual, all the products in the iPhone line-up have received new prices. The following graph shows the current product line-up (US prices before tax) and the…

www.asymco.com

This is hard to read so small, you may want to click the link to see the full thing.
The primary point is what jonsonwax is saying, that Apple designs to fixed price points. They have essentially a floor (at $400 to $450), a mainstream at $600 to $700 (that's what the dots show, up till a few years ago when Apple stopped providing ASP or something similar that could be used to learn that data point), and mainly what they have done since the X is realize that the population that really really cares about their phone is large enough that you might as well cater to them by providing a better phone at a higher price -- NOT a price increase - you can get your iPhone 15 or whatever at the same sort of price as mainstream always was; but if you WANT to spend twice as much for a phone that's a whole lot nicer, Apple will provide that option.


I think we're seeing the same sort of thing play out in M series, though the details are different.
I could be wrong (and jonsonwax, as more on the business side, may have useful counter opinions!) but to my eyes what happened was
- M1/M2 and their Pro/Max/Ultra versions were, apart from the technical elements (growing the number of clusters, figuring out the details of synchronizing power, interrupts etc across two chiplets, and so on) also a probing of the relative demands of M-level compute at different prices

- the resolution of what was learned is going to be an M split somewhat like the Intel i/Xeon split. Mainsteam will be M-class (as in M3 and M4). Enthusiast will be Pro (rebalanced as of the M3).
Separately we will have the "Xeon" line, sold to business (and more precisely sold to the people who run the numbers and see the value of what they are getting compared to the revenue it generates, as opposed to sold to entitled whiners who complain about price and everything else). These will be very very nice compute engines - sold at prices that somewhat reflect their niceness.
Technically I expect the goal here to be how do we cover as wide a dynamic range as possible given that these designs sell in rather smaller quantities. I *think* the answer (for now, things always change...) ties together various elements we've already seen or are rumored
- imagine a chip we call the Max Lite which consists of the SCALABLE elements people care about -- CPU, GPU, memory controller, probably ANE
- imagine a square central "hub" chip which contains the NON-SCALABLE elements - media, display, IO, ISP
- an SKU consists of one, two, four Max Lite's, each connected to one side of a hub chip

This gives us something like Max/Ultra line, but now scalable to an Extreme, and with some area and cost reduction (no duplication of stuff few care about having duplicated, like IO and display support). It also allows for a "server" design (which may only be Apple internal) with a replacement hub chip that drops some of this stuff (eg media, display, ISP, different IO) and allows for different geometry - eg various ways to configure the design to grow to six or more Max Lites.
This also allows for various schemes (some of which I've discussed in previous comments) for substantial increase in installed memory, but again we're only paying the area and power cost for this possibility on the "Xeon" side.

I suspect that (again, somewhat like Xeon) we may also see "skipped" releases of the "Xeon" line. M3 was only consumer, M5 will be only consumer, etc; while M4 and M6 will get the full treatment.

So again I don't expect changes to "established" pricepoints in the Mac space, but I do think we may see new higher price points available to businesses and departments willing to pay; so eg a Studio that offers not just a Max and Ultra config but now also an Extreme (4 Max Lites) config.

 

[Doug S]

My understanding was that N3P was volume production in 2nd half 2024.

The START of mass production, sure. N3E mass production stated in H2 2023 (which turned out to be December) but we didn't see the first product using it ship until May 2024.

Unless there were some rumors that TSMC was pulling in N3P so it would follow N3E by more like six months than the usual twelve, there's no chance Apple can release Macs using N3P in this calendar year. Note that for N3P to follow N3E by a mere six months they would need to have already started N3P mass production earlier this month. That ignores stuff like possibly using risk production wafers (Pro and Max are lower volume, so that's at least possible) but as far as we know Apple hasn't ever done so.

Moreover, the improvements from N3P are minimal, hardly worth delaying the release of new Macs by months to obtain. That sort of discussion might be worth having next year over N3P vs N2, since the improvements will be larger, especially for power.

 

[igor_kavinski]

Moreover, the improvements from N3P are minimal, hardly worth delaying the release of new Macs by months to obtain.

Could be they are delaying for the following reasons:

Thinner laptops (design and validation time would be increased).

Larger Tandem OLED screens.

If not larger OLED displays, then maybe a new MBA 13-inch with Tandem OLED and they are testing this new OLED in the field with the iPad Pro and prefer to wait a few months to see if they might have to scrap plans for that in the M4 MBA model due to real issues encountered by M4 iPad Pro users.

Faster RAM (maybe even faster than 8448 MT/s used by SD Elite X) for which they are building enough inventory.

Sudden pivot towards AI ruined their launch plans and now they will launch later with more focus on AI (maybe with much beefier NPU?)

 

[johnsonwax]

Apple is ALL about segmenting their products in a way that it entices people to upgrade to the next tier. a MacBook Air with higher single-core and multi-core CPU performance than their M3 Pro MacBook Pros is a situation they want to avoid if they can. GPU performance matters a lot less than CPU to most people. There's literally nothing pushing them to release M4 Macs except maybe TSMC trying to get them off N3B so it makes more sense for them to wait until the full lineup is ready.

Sure, but Apple is also all about being in control of their core technology and my thesis here is that M3 kind of ran off into the weeds in terms of yield/cost likely because they pushed TSMC too hard to get 3nm up and running for A17/M3 and chased a process that they couldn't really rein in. Thankfully N3E seems to be doing great.

Apple has a contradiction - yes, they are all about segmenting their products, but they are also all about managing costs and supply chain, and that, I'm suggesting, is currently a little fucked. The question then is if I'm correct (and not the first or only one suggesting this) then do they reconcile that by adherence to the segmentation (which I note, they already broke by putting iPad as the fasted device in the lineup - breaking it in a brand new way) or by rearranging the deck chairs to get away from the yield/cost problem. Not only am I arguing that they would choose the latter, I just offered evidence they've already started doing that - why did the M4 launch in the iPad? Why was their no concurrent step up of A series iPads to A17? The segmentation is already busted. What we observe though, is that Apple appears to be avoiding N3B components at least with the limited information we have so far. I'm backfilling a rationale for why that would be, and if that is what they're doing, the iPhone that would normally move to an N3B A17 as to not further congest the current node, will jump to an A18 to avoid N3B, and that additional 100 million A18 processors all by itself probably prevents further M4 production because like it or not, iPhone is the most important product and everything yields to it.

 

[johnsonwax]

I could be wrong (and jonsonwax, as more on the business side, may have useful counter opinions!) but to my eyes what happened was
- M1/M2 and their Pro/Max/Ultra versions were, apart from the technical elements (growing the number of clusters, figuring out the details of synchronizing power, interrupts etc across two chiplets, and so on) also a probing of the relative demands of M-level compute at different prices

Yep, that's it.

Apple's high end has always been a dogs breakfast. Sometimes it works pretty well, sometimes it's completely missing, sometimes it's 'what the fuck'. I think you can put a bit of that back on Apple struggling to provide Intel based offerings that weren't just repackaging of Dell.

Apple's biggest problem here though is that the job-to-be-done of the Mac is not as robust as the PC market. What's their high-end market? Ok, video editing. Audio, sure. Developers. Uh... LLM hobbyists? This is really the problem they need to work on, and as that comes into greater focus, what that high end needs to look like will similarly come into focus. I mean, I think the single most important thing they've done to secure their high-end sales is ProRes codecs because that is the heart of that market. I did data science on Macs, but that's the sort of thing that you don't really scale your hardware to fit, when things get computationally expensive you run and rent AWS compute and drive it from you MBP. The Mac is fantastic for data science, but it mostly ignores the high end for scalable compute. They're making a better than halfhearted effort on gaming, but still missing the big pieces. You need both the hardware and software to be there. Apple needs to find the software the platform lacks - buy it, make it, but get it.

But yeah, the floor will remain the floor. They'll look for opportunities to go upmarket.

 

[Eug]

Could be they are delaying for the following reasons:

Thinner laptops (design and validation time would be increased).

Larger Tandem OLED screens.

If not larger OLED displays, then maybe a new MBA 13-inch with Tandem OLED and they are testing this new OLED in the field with the iPad Pro and prefer to wait a few months to see if they might have to scrap plans for that in the M4 MBA model due to real issues encountered by M4 iPad Pro users.

There is currently no evidence that the MB Air will get OLED anytime soon. There are predictions from a very reliable leaker that the MB Pro WILL get tandem OLED (or at least the higher end models will), but that it won't be until 2026. IOW, the next MacBook Pros in Q4 2024 or early 2025 will NOT get OLED, so it doesn't have to wait for that.

As for the MB Air, I had previously thought that perhaps its SoC would eventually lag behind the base MB Pro by 1 M generation, but that seems pretty unlikely given that even the iPad Pro gets the latest M series chip. Apple seems intent on having the newest MB Airs getting the latest M series chip, but provides that lower tier MB Air by selling the previous model too.

BTW, some have complained that the tandem OLED in the iPad Pro has a bit of a "grain" effect. AFAIK this is the manifestation of somewhat inconsistent brightness in each pixel (subpixel?) which is most evident on solid grey at very low brightness, in a darkened room. I have specifically looked for this and yes I can see it if I look really closely (like if I'm like 6 inches from the screen or something), but fortunately at my usual viewing distance I don't notice it. I don't know if the MB Pros will have this controlled better or not, but for my usage it's not a significant issue so I don't know if Apple will even bother to address this for the first generation OLED MB Pros.

Otherwise the screens in these iPad Pros are stunning.

 

[trivik12]

Since I have a shiny new account to break in, I'll address this as well.

Apple doesn't increase prices. I know that sounds incredible, but really, they pretty much don't. Apple sets price points and designs to price point. The design team gets a set of requirements and they have latitude in terms of how to hit the price point. Apple may well specify 16GB as a minimum due to the increased compute needs of these learning models, so it's up to the engineers to figure out how to build that and stay at the stated price point (not that it'd be hard, Apples cost on 8GB RAM is what, $8 tops?). Sometimes they'll shift it slightly up or down depending on if there's a particularly good or poor value for money - for a $50 increase we can switch from a higher capacity HDD to a lower capacity SSD and get an order of magnitude improvement on read/write speeds, etc. And then they'll revert the price the following year when SSD prices come down.

Here's my evidence. Go look at the price for the base model iMac:



Go look up what the base price for the iMac was when it was first introduced in 1998:



The iMac hasn't moved more than $100 in price over 26 years. Apple sets a price and then value-adds to justify the price. If they go to 16GB RAM, it'll be $1299. They rely on inflation to make the product more affordable over time, and the even faster rate of technology cost decline to be able to consistently value-add.

Don't confuse how Apple upsells their products with how they set their product strategy. They are doing completely different things. Apple has the Hershey bar and Arizona Ice Tea beat for price stability. That's a quarter century of price stability. I know they have a reputation of always increasing prices - they very rarely do. They will increase the ceiling on what you can upsell to, but the base rarely moves. iPhone has moved up more than anything else. Others have moved down. MacBook Air from $1799 in 2008 to $1099 now.

Didn't they just do it with Ipad Pros. 11" started at 749 or something and with new OLED its starting at 999. 13" went up $300. Even if you increase base storage to 256 its still more expensive than last gen. So if they are upping the specs I expect price increase. 15 pro max increased base storage to 256 and price went up by 100 bucks. I dont see Apple give anything for free.