Why do nonstandard MBP configurations take 3 weeks? They're probably just sitting in a warehouse...

[danwells]

Just a heads up on this. If your BTO configuration craps out, you will have to wait 3 weeks for them to build another one that they can replace it with, assuming they actually have parts. A colleague of mine was shot with this and was down for 6 weeks and had to buy an off the shelf config as a backup. I religiously by off the shelf configs these days - can get them fixed same day (not that I've needed to)

The problem with that, from my perspective as a photographer, is that all the off the shelf configurations have the lowest possible RAM and disk for a given processor (unless the lowest possible is REALLY low, like 8 GB of RAM or 256 GB of storage). In the case of the M3 Max, the only two off the shelf configurations are minimums for the particular sub-variant of the CPU.

I can barely fit all of my applications in 1 TB with none of the images I'm working on. Anything bigger is a CTO.

 

[picpicmac]

If I were Apple, I might still warehouse the North American variations in North America.

IIRC, Apple does have a warehouse in southeast Pennsylvania, to service the market in that part of the country.

There are few products that Apple appears to over-produce. One of them happens to be the M1 Max Mac Studio. OWC (and perhaps others) still have NOS for that.

Apple has a global market, and at least for lower cost items, anything that is not sold at a turnover in models can certainly be sold somewhere in the world.

 

[JayDomK]

The problem with that, from my perspective as a photographer, is that all the off the shelf configurations have the lowest possible RAM and disk for a given processor (unless the lowest possible is REALLY low, like 8 GB of RAM or 256 GB of storage). In the case of the M3 Max, the only two off the shelf configurations are minimums for the particular sub-variant of the CPU.

I can barely fit all of my applications in 1 TB with none of the images I'm working on. Anything bigger is a CTO.

It used to seem like a huge volume to me. But the more applications I started using, the more space was required. 1TB is starting to become obsolete.

 

[AlmightyKang]

The problem with that, from my perspective as a photographer, is that all the off the shelf configurations have the lowest possible RAM and disk for a given processor (unless the lowest possible is REALLY low, like 8 GB of RAM or 256 GB of storage). In the case of the M3 Max, the only two off the shelf configurations are minimums for the particular sub-variant of the CPU.

I can barely fit all of my applications in 1 TB with none of the images I'm working on. Anything bigger is a CTO.

That’s the old “buy two for twice the price” problem domain then.

 

[jlc1978]

Yes, Apple has factored in price drops. But it's free money on the table if Apple does BTO and doesn't stockpile. In exchange, the customer waits a couple weeks. In general, the customers that want a specific config are willing to wait a bit. This is the same practice if you order from Lenovo or Dell.

I agree, and that was my point as well - BTO cuts cost and Apple is not likely to lose those sales.

 

[kevcube]

Second reason is Apple doesn’t stockpile their own M processors either. If they discover a bug that requires a respin of the silicon, they don’t want to end up with a bunch of old M2 chips. Intel and AMD have a few steppings per product.

this is really unlikely. retooling the processes to print silicon is a billion dollar endeavor, and practically doesn't happen. bugs in silicon are fixed in software at the user's expense in the form of performance, and then fixed in hardware in the next generation (or two, because this planning is done so far ahead of time)

 

[danwells]

The one exception to this that has been speculated about is IF (and it's a pretty big if for the reasons Kevcube stated) Apple were to release a second processor named M3 (or Apple might call it M3e or something) in the MacBook Air and iPad Pro, eventually making its way to the iPad Air (and Vision Pro?) as well. It would be pretty much the same processor, but on a revised version of TSMC's 3 nm process (N3e instead of N3b, to get pedantic).

The advantage would be improved yield and possibly power efficiency. The disadvantage would be having to revise the design within a generation. In the case of the M3, it just might be worth it. It would power two or more products that are just about guaranteed to sell a ton of units, and the current 3nm process is known to have yield issues. If Apple was fully satisfied with M3 yield, they would have likely gone ahead and released the high-volume, relatively low cost MacBook Air in time for the holidays. Instead, they went for a prestige product (MacBook Pro) and a niche product in drastic need of an update (iMac).

After a relatively minor performance improvement in the M2 Pro and Max, Apple got their swagger back in the pro laptop market with the M3 Max (while repositioning the M3 Pro closer to the middle between the M3 and the Max), but they didn't move their volume products over . The base M3 leaves core counts untouched, so it's not a massive performance improvement like the M3 Max, but it's respectable, and it should also bring a battery life improvement.

The major reason not to do the volume products right away is that they simply couldn't get enough M3 chips . Maybe the yield issues will work themselves out as TSMC gets more experience with the process, and maybe it's a matter of bringing more lithography units online. Maybe there was capacity taken up by the iPhone 15 Pro launch (the A17 Pro uses the same N3b process as the current) M3 line that is now available, but they might also be waiting for a higher-yield version of the 3nm process.

A respin of the base M3 on the newer N3e process would have another advantage. Apple's processors are closely enough related that it would mean a lot of the work had been done to respin the A17 pro. There's the processor for the iPhone 16 base models... Sure, they'd be introducing a node in the MacBook Air instead of the iPhone, which they don't normally do - but it's a minor node whose major advantage is Apple-facing (higher yield), rather than customer-facing.

 

[JPack]

this is really unlikely. retooling the processes to print silicon is a billion dollar endeavor, and practically doesn't happen. bugs in silicon are fixed in software at the user's expense in the form of performance, and then fixed in hardware in the next generation (or two, because this planning is done so far ahead of time)

Making a new mask for a new stepping doesn’t cost a billion dollars. And we already know it happens. Apple says there are different versions of A12 and A13 chips depending on when consumers bought their iPhone.

Secure Enclave

The Secure Enclave is a dedicated secure subsystem in the latest versions of iPhone, iPad, Mac, Apple TV, Apple Watch, and HomePod.

support.apple.com

What Does Stepping of Intel® Processors Mean?

Explains "Stepping" field for the Intel® Processor.

www.intel.com

 

[ondioline]

A clear piece of evidence that BTO systems are custom assembled and not bulk manufactured and stored in a warehouse (lol) was the 27 inch iMac. Depending on your config it would either come from China or Ireland. The Ireland iMacs were almost always the relatively rarer VESA/i9/etc varieties.

 

[JPack]

The one exception to this that has been speculated about is IF (and it's a pretty big if for the reasons Kevcube stated) Apple were to release a second processor named M3 (or Apple might call it M3e or something) in the MacBook Air and iPad Pro, eventually making its way to the iPad Air (and Vision Pro?) as well. It would be pretty much the same processor, but on a revised version of TSMC's 3 nm process (N3e instead of N3b, to get pedantic).

The advantage would be improved yield and possibly power efficiency. The disadvantage would be having to revise the design within a generation. In the case of the M3, it just might be worth it. It would power two or more products that are just about guaranteed to sell a ton of units, and the current 3nm process is known to have yield issues. If Apple was fully satisfied with M3 yield, they would have likely gone ahead and released the high-volume, relatively low cost MacBook Air in time for the holidays. Instead, they went for a prestige product (MacBook Pro) and a niche product in drastic need of an update (iMac).

After a relatively minor performance improvement in the M2 Pro and Max, Apple got their swagger back in the pro laptop market with the M3 Max (while repositioning the M3 Pro closer to the middle between the M3 and the Max), but they didn't move their volume products over . The base M3 leaves core counts untouched, so it's not a massive performance improvement like the M3 Max, but it's respectable, and it should also bring a battery life improvement.

The major reason not to do the volume products right away is that they simply couldn't get enough M3 chips . Maybe the yield issues will work themselves out as TSMC gets more experience with the process, and maybe it's a matter of bringing more lithography units online. Maybe there was capacity taken up by the iPhone 15 Pro launch (the A17 Pro uses the same N3b process as the current) M3 line that is now available, but they might also be waiting for a higher-yield version of the 3nm process.

A respin of the base M3 on the newer N3e process would have another advantage. Apple's processors are closely enough related that it would mean a lot of the work had been done to respin the A17 pro. There's the processor for the iPhone 16 base models... Sure, they'd be introducing a node in the MacBook Air instead of the iPhone, which they don't normally do - but it's a minor node whose major advantage is Apple-facing (higher yield), rather than customer-facing.

There is no "respinning" an N3B product.

N3E uses different design rules compared to N3B. SRAM and logic density are also different. That's why other TSMC customers like Qualcomm and MediaTek are waiting for N3E because it's incompatible with existing rules.

 

[danwells]

There is no "respinning" an N3B product.

N3E uses different design rules compared to N3B. SRAM and logic density are also different. That's why other TSMC customers like Qualcomm and MediaTek are waiting for N3E because it's incompatible with existing rules.

Call it what you will - valid point that it is more than a respin. There is some chance of a mid-cycle move of M3 (or a very M3-like chip) to N3E, and I suspect a larger chance of the base iPhone 16 using a chip that is a lot like A17 Pro, but built on N3E. Apple has to do the work anyway because of the iPhone 16 base models (and probably the M3-class iPad Airs) - they could use that same work to produce an M3-ish chip (that they would possibly include in the M3 series) for laptops...

 

[JPack]

Call it what you will - valid point that it is more than a respin. There is some chance of a mid-cycle move of M3 (or a very M3-like chip) to N3E, and I suspect a larger chance of the base iPhone 16 using a chip that is a lot like A17 Pro, but built on N3E. Apple has to do the work anyway because of the iPhone 16 base models (and probably the M3-class iPad Airs) - they could use that same work to produce an M3-ish chip (that they would possibly include in the M3 series) for laptops...

While there's a chance for a refreshed M3, I think it's very unlikely. The iPad Pro and MacBook Air coming next month will be using M3 already.

N3E differs from N3B significantly. This means Apple will need to redesign the chip with a different cache to logic ratio in mind (due to N3E's lower SRAM density) and adjust frequency. M3 is also missing Thunderbolt 5 among other features. All of this means there's unlikely to be a reheated M3. It makes more sense to launch a M4 in late 2024 or 2025.

All of the above applies to iPhone 16. The A17 Pro is a massive chip with a USB3 controller, Pro display engine, and too many GPU cores. It doesn't make sense to cut down on those features and reuse old P- and E-cores. That is effectively a new chip design. That's why current rumors indicate iPhone 16 will use A18 and not A17 Lite.

 

[danwells]

A clear piece of evidence that BTO systems are custom assembled and not bulk manufactured and stored in a warehouse (lol) was the 27 inch iMac. Depending on your config it would either come from China or Ireland. The Ireland iMacs were almost always the relatively rarer VESA/i9/etc varieties.

View attachment 2348703

iMacs, especially the larger models, certainly were CTOs in the classic sense - the 27" iMac had thousands of configurations, and at least three modular internal parts that could be interchanged when the machine was 80% complete (CPU, RAM, storage). I believe some models also had the GPU on a daughterboard. I don't know whether the gigabit vs. 10 Gb Ethernet choice in the last generation was a daughterboard, or whether that was two different motherboards?

That's exactly like the classic Dell desktop CTO model (or what every smaller system builder like Puget Systems does). There are a bunch of interchangeable parts, order one from each menu (or maybe several, in the case of RAM or multiple drives/GPUs). Then somebody snaps them together... Sometimes it's three or four parts like an iMac, sometimes it's more like a traditional Mac Pro with six or seven parts, but there are a bunch of parts that go into a barebones chassis.

What makes the current crop of Macs different is that they are almost all single-board with no modular components. There is one part that makes a $2499 base 16" MBP different from a $7199 fully upgraded model. One logic board has a M3 Pro, 18GB of RAM and 512 GB of storage, while the other has a 16/40 M3 Max, 128 GB of RAM and 8 TM of storage, but every one of those things is fully integrated on the logic board. Every other part is the same. I had forgotten about the keyboards, because so many of them are US or International English, but that DOES introduce another set of variations.

While there's a chance for a refreshed M3, I think it's very unlikely. The iPad Pro and MacBook Air coming next month will be using M3 already.

N3E differs from N3B significantly. This means Apple will need to redesign the chip with a different cache to logic ratio in mind (due to N3E's lower SRAM density) and adjust frequency. M3 is also missing Thunderbolt 5 among other features. All of this means there's unlikely to be a reheated M3. It makes more sense to launch a M4 in late 2024 or 2025.

All of the above applies to iPhone 16. The A17 Pro is a massive chip with a USB3 controller, Pro display engine, and too many GPU cores. It doesn't make sense to cut down on those features and reuse old P- and E-cores. That is effectively a new chip design. That's why current rumors indicate iPhone 16 will use A18 and not A17 Lite

Your thought is that there will be an A18 and an A18 Pro? More on the concept of M3, M3 Pro, M3 Max than the current trend of using the previous year's chip in the non-pro model. If that's the case (and it makes sense), I'd see an M3 remake on N3E as quite unlikely.

They'd go M4 as soon as possible (and including the Airs, not wanting the cheaper computers on N3B with lower yields for any longer than they had to) after the iPhone. I'd be absolutely shocked to see a "base M4 surprise" BEFORE the iPhone (e.g. the Airs that about to come out having a processor tagged "M4") - the rule that anything they count as a generation starts with the iPhone seems inviolable. I could see these upcoming Airs being replaced very quickly, like they did with the M2 Pro/Max. A March release of the M3 Airs, followed by an October/November release of the N3E M4 Airs seems at least possible.

This idea of two iPhone chips (that are designed as different models, rather than being a year apart) has an interesting effect on the iPad line. Right now, Apple has an iPad problem. They're trying to release the iPad Pro and iPad Air simultaneously. Giving both the M3 is tricky - sure, they can put a fancy OLED screen on the Pro, and maybe that's enough. Unless they can somehow cram the M3 PRO into the iPad Pro (really unlikely - people are saying that it won't even fit in the 15" MacBook Air as an upgrade chip possibility), they don't have a good way to differentiate the two on a performance basis.

If they stuck the A17 Pro into the iPad Air, leaving the M3 as iPad Pro only, they move the problem down the line. They don't have a good chip for the base iPad and the Mini (they either have to use the A16, which will be almost two years old when they update the base iPads, or they have to use the same chip that went in the iPad Air only 5 months earlier).

Having two same-year phone chips gives them a better selection of iPad chips (maybe not this generation but going forward). The iPad Pro gets the base M-series chip (the Vision Pro probably does as well), the iPad Air gets the chip from the Pro phones, and the base iPad and the Mini get the chip from the non-Pro phones. Each gets the current version of that chip at the time of their release.

 

[JPack]

Your thought is that there will be an A18 and an A18 Pro? More on the concept of M3, M3 Pro, M3 Max than the current trend of using the previous year's chip in the non-pro model. If that's the case (and it makes sense), I'd see an M3 remake on N3E as quite unlikely.

They'd go M4 as soon as possible (and including the Airs, not wanting the cheaper computers on N3B with lower yields for any longer than they had to) after the iPhone. I'd be absolutely shocked to see a "base M4 surprise" BEFORE the iPhone (e.g. the Airs that about to come out having a processor tagged "M4") - the rule that anything they count as a generation starts with the iPhone seems inviolable. I could see these upcoming Airs being replaced very quickly, like they did with the M2 Pro/Max. A March release of the M3 Airs, followed by an October/November release of the N3E M4 Airs seems at least possible.

A18 will eventually reach iPhone SE, low-cost iPad, Apple TV, and maybe HomePod. It makes sense to design a dedicated cost-optimized A18, A19, etc. without fusing off sections of A17/A18 Pro and paying for unused transistors.

Looking to M4, I suspect it's a 2025 product. Apple needs to recoup the the billion dollar NRE costs of M3 but Mac sales haven't been spectacular. Putting an M4 into MacBook early isn't going to drive sales much, not without a chassis or form factor redesign. As a result, I suspect M3 will remain for about 18-months to match a form factor refresh.

This idea of two iPhone chips (that are designed as different models, rather than being a year apart) has an interesting effect on the iPad line. Right now, Apple has an iPad problem. They're trying to release the iPad Pro and iPad Air simultaneously. Giving both the M3 is tricky - sure, they can put a fancy OLED screen on the Pro, and maybe that's enough. Unless they can somehow cram the M3 PRO into the iPad Pro (really unlikely - people are saying that it won't even fit in the 15" MacBook Air as an upgrade chip possibility), they don't have a good way to differentiate the two on a performance basis.

If they stuck the A17 Pro into the iPad Air, leaving the M3 as iPad Pro only, they move the problem down the line. They don't have a good chip for the base iPad and the Mini (they either have to use the A16, which will be almost two years old when they update the base iPads, or they have to use the same chip that went in the iPad Air only 5 months earlier).

Having two same-year phone chips gives them a better selection of iPad chips (maybe not this generation but going forward). The iPad Pro gets the base M-series chip (the Vision Pro probably does as well), the iPad Air gets the chip from the Pro phones, and the base iPad and the Mini get the chip from the non-Pro phones. Each gets the current version of that chip at the time of their release.

Apple's current strategy is to offer the latest M-series chips to iPad Pro and give iPad Air n-1. This means iPad Air should receive M2 this spring. Combined with the difference in Face ID and rear cameras, this should put enough separation between iPad Pro and iPad Air.

 

[danwells]

Looking to M4, I suspect it's a 2025 product. Apple needs to recoup the the billion dollar NRE costs of M3 but Mac sales haven't been spectacular. Putting an M4 into MacBook early isn't going to drive sales much, not without a chassis or form factor redesign. As a result, I suspect M3 will remain for about 18-months to match a form factor refresh.

The problem with that is that it keeps the low-yield N3B process around for a long time. I agree completely that chips don't really drive MacBook Air (or iMac) sales - the average MBA customer wants a laptop that Just Works. They're a Word/PowerPoint (maybe)/Mail/Safari user who watches some movies and makes some FaceTime calls. They may be motivated by a new chassis or new colors, or they may just buy what's at the Apple Store when they need a computer (they have an iPhone, or have owned a previous Mac, and want a Mac).

The MacBook Pro/Mac Studio customer is a little different (they do ask for chips by name, and they didn't love the M2 line's performance jump over the M1 line)- but the M3 Max is a great upgrade, and the M3 Ultra is going to be a beast. The people who want those fire breathers know why. The M3 Pro is a funny chip - it makes a great deal of sense in rationalizing the line (and it offers upgrades in performance and RAM without the expense or heat of a Max) - but its drop in P-cores makse people unlikely to request it by name. It's a great business laptop chip, but many people buying business laptops don't know what's in them...

Apple's best reason for getting M4 (and, by extension, N3E) out there isn't driving sales (unless they add some new capability I'm not anticipating) - it's cutting costs. There was an article saying that TSMC doesn't charge Apple for faulty chips (an unusual arrangement - most customers buying custom chips have to buy by the wafer, so the foundry passes on the cost of defects). Even so, TSMC isn't stupid - they have to be charging Apple more when yields are low, unless they gave Apple a short-term sweetheart deal as a launch customer. If so, they're going to be changing that as soon as they can. However you slice it, low-yield chips are going to cost more than higher-yield chips, so (unless N3B yield improves), Apple wants to move to N3E as soon as they can.

 

[dmccloud]

The problem with that is that it keeps the low-yield N3B process around for a long time. I agree completely that chips don't really drive MacBook Air (or iMac) sales - the average MBA customer wants a laptop that Just Works. They're a Word/PowerPoint (maybe)/Mail/Safari user who watches some movies and makes some FaceTime calls. They may be motivated by a new chassis or new colors, or they may just buy what's at the Apple Store when they need a computer (they have an iPhone, or have owned a previous Mac, and want a Mac).

The MacBook Pro/Mac Studio customer is a little different (they do ask for chips by name, and they didn't love the M2 line's performance jump over the M1 line)- but the M3 Max is a great upgrade, and the M3 Ultra is going to be a beast. The people who want those fire breathers know why. The M3 Pro is a funny chip - it makes a great deal of sense in rationalizing the line (and it offers upgrades in performance and RAM without the expense or heat of a Max) - but its drop in P-cores makse people unlikely to request it by name. It's a great business laptop chip, but many people buying business laptops don't know what's in them...

Apple's best reason for getting M4 (and, by extension, N3E) out there isn't driving sales (unless they add some new capability I'm not anticipating) - it's cutting costs. There was an article saying that TSMC doesn't charge Apple for faulty chips (an unusual arrangement - most customers buying custom chips have to buy by the wafer, so the foundry passes on the cost of defects). Even so, TSMC isn't stupid - they have to be charging Apple more when yields are low, unless they gave Apple a short-term sweetheart deal as a launch customer. If so, they're going to be changing that as soon as they can. However you slice it, low-yield chips are going to cost more than higher-yield chips, so (unless N3B yield improves), Apple wants to move to N3E as soon as they can.

The "low" yield is neither a permanent issue nor a static one. Any new process will have low yields at first, then increase as the processes and methodologies mature - it has never been a static figure. N3E will have lower yields at the beginning for the same reason, as does Intel every time they move process nodes (AMD is a bit of an outlier here as they use TSMC like Apple, but usually are 1-2 nodes behind). Even now, N3B yields have increased to the point at which an iPad refresh is imminent. Regarding TSMC, yields one year ago were around 55%, and now TSMC is openly aiming to break the 80% mark. Last December, TSMCs 3nm yields were estimated to be between 60-80%, so the yields have been improving steadily over the last year. Assuming TSMC reaches this goal, it could result in doubling their production capacity.