M4+ Chip Generation - Speculation Megathread [MERGED]

[tenthousandthings]

Ah I see. Is this die shot of the iPad Pro M4? And is it thought that the M4 that goes into the iPad Pro is likely to be the very same one that will go into the rest of the lineup later?

I ask about this in part because I am interested in getting an Apple Studio Display. People always say get what you need when you need it but in this case, it would be an upgrade and is not a must have, and if a better one may come out in a few months I'll wait. But if it likely won't come before the M5 desktops are out, then I might as well grab the current one....

Well, the answer to my question seems to be “No, this is not what TB5 support would look like,” so that’s not a good sign for new/updated displays during the M4 Mini/Studio/Pro event, whenever and whatever it will be.

That said, IMHO, the displays are a product weak point. Apple has replaced the 5K iMac/iMac Pro with the 5K Studio Display, but they haven’t yet followed through on the promise of the Mac Studio (an expanded range of Apple displays) beyond the Pro Display XDR.

 

[Confused-User]

That said, IMHO, the displays are a product weak point. Apple has replaced the 5K iMac/iMac Pro with the 5K Studio Display, but they haven’t yet followed through on the promise of the Mac Studio (an expanded range of Apple displays) beyond the Pro Display XDR.

There is almost no demand for 8k. Even among Apple's pro customers demand is likely pretty low. I doubt they have any plans to do an 8k monitor this year or next. Bringing down the cost of 6k and broadening that customer base is likely more interesting.

I'd also love to see them push the 5k under $1000, but I doubt they'll do that. It's too bad they don't seem to have much interest in serving the $500-$1200 market, but I think that's because they insist on retina and high quality and I don't think they see a niche for that at a lower price point. Maybe 4.5k 24", but I doubt it.

 

[Christopher Pelham]

I'd be happy to see the current 27" 5K simply updated to 120Mhz or 144Mhz refresh rate and a better camera. A 32" 5k or 6k option might be fairly popular with content creators as well.

 

[tenthousandthings]

There is almost no demand for 8k. Even among Apple's pro customers demand is likely pretty low. I doubt they have any plans to do an 8k monitor this year or next. Bringing down the cost of 6k and broadening that customer base is likely more interesting.

I'd also love to see them push the 5k under $1000, but I doubt they'll do that. It's too bad they don't seem to have much interest in serving the $500-$1200 market, but I think that's because they insist on retina and high quality and I don't think they see a niche for that at a lower price point. Maybe 4.5k 24", but I doubt it.

Yes, as I'm sure you know, Dell's 8K 32" hasn't changed since 2017. Nobody else has tried. M2 Studio supports it and it works well (using two TB ports), not sure about other Macs. So if you want 8K with "Liquid Retina" pixel density, you can get it now. Apple has no incentive here.

My own fantasyland would be as follows:

Studio Display Retina 5K 27" (5120x2660)
Studio Display Retina 6K 32" (6144x3456) [This is the Dell "6K" resolution.]

Pro Display XDR Retina 34" (6400x3600)

 

[EugW]

Yes, as I'm sure you know, Dell's 8K 32" hasn't changed since 2017. Nobody else has tried. M2 Studio supports it and it works well (using two TB ports), not sure about other Macs. So if you want 8K with "Liquid Retina" pixel density, you can get it now. Apple has no incentive here.

My own fantasyland would be as follows:

Studio Display Retina 5K 27" (5120x2660)
Studio Display Retina 6K 32" (6144x3456) [This is the Dell "6K" resolution.]

Pro Display XDR Retina 34" (6400x3600)

I still would much rather have a 29-30” 5K.

 

[Chuckeee]

Yes, as I'm sure you know, Dell's 8K 32" hasn't changed since 2017. Nobody else has tried. M2 Studio supports it and it works well (using two TB ports), not sure about other Macs. So if you want 8K with "Liquid Retina" pixel density, you can get it now. Apple has no incentive here.

My own fantasyland would be as follows:

Studio Display Retina 5K 27" (5120x2660)
Studio Display Retina 6K 32" (6144x3456) [This is the Dell "6K" resolution.]

Pro Display XDR Retina 34" (6400x3600)

I was under the impression that a resurgence for 8K displays would go along with TB5

 

[Antony Newman]

+) 8k acquisition - and 4→6k screen.
+) Use of Tandem OLED to get HDR would be useful for editing
+) P3 (or trowards Rec2020)

If the Tandem OLED uses 1/3 the power - it could get reasonable HDR displays without cooking the viewer.

 

[tenthousandthings]

I still would much rather have a 29-30” 5K.

There is a natural Retina resolution for 30" at 5760x3240 (3x 1080p). [4K is 2x 1080p] Definitely a good possibility, and an improvement over the 27" 5K.

 

[tenthousandthings]

I was under the impression that a resurgence for 8K displays would go along with TB5

My favorite in that realm is a 42" Liquid Retina 9600x5400 (5x 1080p) [8K is 4x 1080p], round up and call it 10K? Unless I’m mistaken, Thunderbolt 5 could drive it with a single cable.

 

[Chuckeee]

My favorite in that realm is a 42" Liquid Retina 9600x5400 (5x 1080p) [8K is 4x 1080p], round up and call it 10K? Unless I’m mistaken, Thunderbolt 5 could drive it with a single cable.

All depends on frame rate

 

[tenthousandthings]

To get this thread back on topic (says the person responsible for sending it off topic), there have been two rumors of late that seem relevant.

The first is the indicators (via model identifiers, plus a rumor) that iPhone 16 and 16 Pro will share an SoC generation, likely A18 and A18 Pro, or maybe just A18. The rumor is that A18 will use the A17 Pro architecture, moved to the N3E process node. This is interesting because it could help explain the discrepancy between how Apple has reported the A17 Pro specifications versus the M3 family. For A14/M1 and A15/M2, the neural engine numbers match, but not for A17/M3. Here's my theory about what this means: Apple made changes late in the development process for the A17, after the M3 was put to bed. Changes that would result in the last-minute addition of INT8 support for the A17 Pro (thus explaining the discrepancy), but more importantly, changes that allow the design to move directly from N3 (a.k.a. N3B) to N3E.

The other is a rumor reported by Economic Daily (經濟日報). This is a Taiwan-based outlet that has been around since 1967 (not to be confused with the state-owned mainland Chinese propaganda outlet of the same name). The rumor is that Apple is using TSMC's SoIC advanced packaging technology for the next generation of the M series, and even the A series can't be ruled out:

台積電 SoIC 傳再增大客戶2025放量 載板夥伴欣興同沾光 | 產業熱點 | 產業 | 經濟日報

台積電（2330）先進封裝平台的SoIC傳出再增添重量級客戶蘋果導入使用，預計2025年放量，若成真將成為超微（AMD）...

money.udn.com

Apple M5 Chip's Dual-Use Design Will Power Future Macs and AI Servers

Apple will reportedly use a more advanced SoIC packaging technology for its M5 chips, as part of a two-pronged strategy to meet its growing need for...

www.macrumors.com

I don't have access to the DigiTimes story about the Economic Daily piece, but judging from the MacRumors summary, DigiTimes seems to be adding details, presumably known information about the current state of SoIC development.

In addition, TSMC itself has recently highlighted SoIC (System-on-Integrated-Chips), mentioning it in the press release after their technology symposium in April. So what do we think? Is this plausible? What do we know about SoIC, either SoIC-X or SoIC-P, and SiP (System-in-Package) in general?

 

[Confused-User]

To get this thread back on topic (says the person responsible for sending it off topic), there have been two rumors of late that seem relevant.

The first is the indicators (via model identifiers, plus a rumor) that iPhone 16 and 16 Pro will share an SoC generation, likely A18 and A18 Pro, or maybe just A18. The rumor is that A18 will use the A17 Pro architecture, moved to the N3E process node. This is interesting because it could help explain the discrepancy between how Apple has reported the A17 Pro specifications versus the M3 family. For A14/M1 and A15/M2, the neural engine numbers match, but not for A17/M3. Here's my theory about what this means: Apple made changes late in the development process for the A17, after the M3 was put to bed. Changes that would result in the last-minute addition of INT8 support for the A17 Pro (thus explaining the discrepancy), but more importantly, changes that allow the design to move directly from N3 (a.k.a. N3B) to N3E.

This is as close to being impossible as you can get, without actually being impossible. I would give you 100 to 1 odds against it being true, and not worry about losing.

They are not going to reimplement an A17 P core on N3E when they've got a perfectly good M4 N3E P core already done. That's just lunacy.

What you're seeing is simply an indication that as their design team matures they're more comfortable plugging together different generations of various cores, rather than moving everything in lockstep, depending on when particular cores get done.

 

[crazy dave]

To get this thread back on topic (says the person responsible for sending it off topic), there have been two rumors of late that seem relevant.

The first is the indicators (via model identifiers, plus a rumor) that iPhone 16 and 16 Pro will share an SoC generation, likely A18 and A18 Pro, or maybe just A18. The rumor is that A18 will use the A17 Pro architecture, moved to the N3E process node. This is interesting because it could help explain the discrepancy between how Apple has reported the A17 Pro specifications versus the M3 family. For A14/M1 and A15/M2, the neural engine numbers match, but not for A17/M3. Here's my theory about what this means: Apple made changes late in the development process for the A17, after the M3 was put to bed. Changes that would result in the last-minute addition of INT8 support for the A17 Pro (thus explaining the discrepancy), but more importantly, changes that allow the design to move directly from N3 (a.k.a. N3B) to N3E.

Perhaps I missed something but I was under the impression that the M3 NPU was identical to the A17 NPU, it was just the reporting units that were different? As in the M3 was reported using the old FP16 metric while the A17 was reported using the now maybe industry standard INT8 throughput, but they are actually the same NPU. The M4 NPU was then reported with INT8.

This is as close to being impossible as you can get, without actually being impossible. I would give you 100 to 1 odds against it being true, and not worry about losing.

They are not going to reimplement an A17 P core on N3E when they've got a perfectly good M4 N3E P core already done. That's just lunacy.

What you're seeing is simply an indication that as their design team matures they're more comfortable plugging together different generations of various cores, rather than moving everything in lockstep, depending on when particular cores get done.

Regardless even if the M3 NPU was the old one, what @Confused-User said still holds, there's no reason to port the M3/A17 P-core over to the N3E process when the M4 P-core exists. @tenthousandthings Who is the source for the rumor that the A18 was to use the A17's P-core?

 

[mr_roboto]

What you're seeing is simply an indication that as their design team matures they're more comfortable plugging together different generations of various cores, rather than moving everything in lockstep, depending on when particular cores get done.

They're even starting to do custom versions of some cores for niche (in terms of units sold) products. M3 Max E cores have three ASIMD/FP execution units, while M3 and M3 Pro E cores have only two. This is not speculation, or even reverse engineered info, it's documented by Apple in the Apple Silicon CPU Optimization Guide they published a few months ago.

So while it's usually a good guess that all members of the same generation will have the same cores as building blocks, it's no longer guaranteed. They may be customizing them for specific projects now.

(As to why they would mess around with the E cores on M3 Max? My guess is that they wanted them to contribute more to heavily multi-threaded math-intensive code. M3 Max is essentially Apple's workstation chip and some workstation users do a lot of CPU-based math.)

 

[crazy dave]

They're even starting to do custom versions of some cores for niche (in terms of units sold) products. M3 Max E cores have three ASIMD/FP execution units, while M3 and M3 Pro E cores have only two. This is not speculation, or even reverse engineered info, it's documented by Apple in the Apple Silicon CPU Optimization Guide they published a few months ago.

So while it's usually a good guess that all members of the same generation will have the same cores as building blocks, it's no longer guaranteed. They may be customizing them for specific projects now.

(As to why they would mess around with the E cores on M3 Max? My guess is that they wanted them to contribute more to heavily multi-threaded math-intensive code. M3 Max is essentially Apple's workstation chip and some workstation users do a lot of CPU-based math.)

No kidding! I missed that, I thought it was three on all?

 

[mr_roboto]

No kidding! I missed that, I thought it was three on all?

See table A.7 (ASIMD&FP Execution Class Availability: E Core). It's a little confusing because in the version of the doc I downloaded, the table's broken across two pages, and the break splits the first row. That first row covers all E cores except M3 Max, the second row describes the M3 Max E core.

Only M3 Max E cores have the third ASIMD/FP execution unit. M3 Max also enhances ASIMD/FP unit #1 to handle slightly more instruction types.

 

[tenthousandthings]

Regardless even if the M3 NPU was the old one, what @Confused-User said still holds, there's no reason to port the M3/A17 P-core over to the N3E process when the M4 P-core exists. @tenthousandthings Who is the source for the rumor that the A18 was to use the A17's P-core?

Sorry, I didn't have time this morning to properly source that, and I'd forgotten about the MacRumors error noted below. So I assumed you all had probably seen it, and would know what I was referring to. The source is a CTEE (工商時報) report about iPhone 16, here:

https://www.ctee.com.tw/news/20240701700050-430501

MacRumors covered this but apparently they misunderstood with regard to the part of the CTEE report that says: "...the A18 will use continue the original A17 Pro design with the process node changed to N3E, while the A18 Pro will have a new design..."

I don't know if the MacRumors mistake (saying that A17 Pro was produced on N3E) was an auto-translate error, or what, but it's obviously wrong, since N3E didn't go into volume production until after the iPhone 15 Pro shipped:

Apple Boosts A18 Chip Orders in Anticipation of High iPhone 16 Demand

Apple is said to have upped its order of next-generation chips from TSMC to between 90 million and 100 million units, following heightened demand...

www.macrumors.com

 

[TigeRick]

Sorry, I didn't have time this morning to properly source that, and I'd forgotten about the MacRumors error noted below. So I assumed you all had probably seen it, and would know what I was referring to. The source is a CTEE (工商時報) report about iPhone 16, here:

https://www.ctee.com.tw/news/20240701700050-430501

MacRumors covered this but apparently they misunderstood with regard to the part of the CTEE report that says: "...the A18 will use the original A17 Pro design with the process node changed to N3E, while the A18 Pro will have a new design..."

I don't know if the MacRumors mistake (saying that A17 Pro was produced on N3E) was an auto-translate error, or what, but it's obviously wrong, since N3E didn't go into volume production until after the iPhone 15 Pro shipped:

Apple Boosts A18 Chip Orders in Anticipation of High iPhone 16 Demand

Apple is said to have upped its order of next-generation chips from TSMC to between 90 million and 100 million units, following heightened demand...

www.macrumors.com

Yep, here is the translated version (I am Chinese) with my comment below:

• Upcoming iPhone 16 and 16 Plus will use A18 which offer similar features as A17 Pro (with 8GB LPDDR5) but fabbed by N3E process. It makes sense to me cause A18 will be getting speed bump from current A16 with 6GB of RAM. Apple would most likely upgrade the GPU to include RT cores as well. Of course, Apple might cut off certain high end features like USB3 to maintain product segmentation.
• Upcoming iPhone 16 Pro series will feature A18 Pro SoC with all-new design (that's the article words but I don't think so). The die size would be 15-20% larger than A18 with more GPU cores and NPU. To me, A18 Pro's P-core should offer similar features as M4's P-core.
• All four iPhone 16 will feature 8GB of RAM. Do you think Apple will offer same amount of RAM for all iPhone 16 lineup? That's why I speculated Apple will bump the amount of RAM to 12GB LPDDR6 for iPhone 16 Pro series...we shall see in two months time.

 

[crazy dave]

Sorry, I didn't have time this morning to properly source that, and I'd forgotten about the MacRumors error noted below. So I assumed you all had probably seen it, and would know what I was referring to. The source is a CTEE (工商時報) report about iPhone 16, here:

https://www.ctee.com.tw/news/20240701700050-430501

MacRumors covered this but apparently they misunderstood with regard to the part of the CTEE report that says: "...the A18 will use the original A17 Pro design with the process node changed to N3E, while the A18 Pro will have a new design..."

I don't know if the MacRumors mistake (saying that A17 Pro was produced on N3E) was an auto-translate error, or what, but it's obviously wrong, since N3E didn't go into volume production until after the iPhone 15 Pro shipped:

Apple Boosts A18 Chip Orders in Anticipation of High iPhone 16 Demand

Apple is said to have upped its order of next-generation chips from TSMC to between 90 million and 100 million units, following heightened demand...

www.macrumors.com

Thanks for the source. It seems to me that the Macrumors correctly translates that "A18 will be the A17 design but ported to N3E" in a later paragraph but got confused in an earlier one:

Last year, the iPhone 15 and iPhone 15 Plus were equipped with the A16 Bionic chip – the same chip that was in the iPhone 14 Pro models – whereas the iPhone 15 Pro and iPhone 15 Pro Max featured an A17 Pro chip manufactured using TSMC's second-generation 3nm process, also known as "N3E."

vs

This year, all four models in the iPhone 16 lineup are expected to use A18-branded chipsmanufactured using the N3E process. Despite switching to N3E, the A18 chip in the iPhone 16 and iPhone 16 Plus is likely to adopt the A17 Pro design that was used in the iPhone 15 Pro models.

Beyond the mistranslation issues I wonder if this is a misunderstanding of what part of the A17Pro vs A18Pro "design" is referring to? For instance a "new design for A18Pro" could imply a new SOC/cluster layout like 2+6 in the A18Pro but the A18 regular will be a binned variant with a 2+4 layout, which would be the same as the A17Pro. Something like that would also fit the prediction but not require porting the A17Pro's cores over to N3E (which isn't even cell compatible with A17's N3B, so that's quite a bit of time, effort, and money). I'm just guessing here. But spending money to port the older cores over to a new node just doesn't make any sense to me - not in this case anyway where Apple already has new core designs on the new node.

 

[mr_roboto]

See table A.7 (ASIMD&FP Execution Class Availability: E Core). It's a little confusing because in the version of the doc I downloaded, the table's broken across two pages, and the break splits the first row. That first row covers all E cores except M3 Max, the second row describes the M3 Max E core.

Only M3 Max E cores have the third ASIMD/FP execution unit. M3 Max also enhances ASIMD/FP unit #1 to handle slightly more instruction types.

By the way, for anyone reading this wondering about whether the M4 E core is also better at SIMD and FP... Apple published this before announcing M4. It only covers M1, M2, M3, A14, A15, and A16. No way to predict when they'll add information about M4 or A17. Hopefully this is something they're going to keep reasonably current, but it did take them until early this year to publish the very first version of the document at all, so who knows.

 

[tenthousandthings]

Thanks for the source. It seems to me that the Macrumors correctly translates that "A18 will be the A17 design but ported to N3E" in a later paragraph but got confused in an earlier one: [...]

Beyond the mistranslation issues I wonder if this is a misunderstanding of what part of the A17Pro vs A18Pro "design" is referring to? For instance a "new design for A18Pro" could imply a new SOC/cluster layout like 2+6 in the A18Pro but the A18 regular will be a binned variant with a 2+4 layout, which would be the same as the A17Pro. Something like that would also fit the prediction but not require porting the A17Pro's cores over to N3E (which isn't even cell compatible with A17's N3B, so that's quite a bit of time, effort, and money). I'm just guessing here. But spending money to port the older cores over to a new node just doesn't make any sense to me - not in this case anyway where Apple already has new core designs on the new node.

Hmm, yes, I see there is an earlier mention of N3E in the first paragraph of the CTEE report, which must be the source of the mistranslation in MacRumors. The report just says all iPhone 16 processors will be on N3E.

The Safari Chinese-to-English translation of that page is not bad, and conveys the statement in question accurately: "...the A18 processor, continuing the original A17 Pro processor design, but changing to the TSMC N3E process..."

The word translated as "design" is 設計 shèjì, meaning a design in the sense of a plan or a layout. My knowledge of technical-scientific Chinese usage is not sufficient to read anything more into it, but for what it's worth my initial reading was "the original A17 Pro processor architecture." But honestly I'm not altogether clear on how "architecture" is used in English computer-science jargon, let alone Chinese, so I defaulted back to the more generic "design." I didn't want to read something into it that wasn't there.

 

[Confused-User]

Beyond the mistranslation issues I wonder if this is a misunderstanding of what part of the A17Pro vs A18Pro "design" is referring to? For instance a "new design for A18Pro" could imply a new SOC/cluster layout like 2+6 in the A18Pro but the A18 regular will be a binned variant with a 2+4 layout, which would be the same as the A17Pro. Something like that would also fit the prediction but not require porting the A17Pro's cores over to N3E (which isn't even cell compatible with A17's N3B, so that's quite a bit of time, effort, and money). I'm just guessing here. But spending money to port the older cores over to a new node just doesn't make any sense to me - not in this case anyway where Apple already has new core designs on the new node.

That would be a strange way to write about it, but it's the only reasonable explanation I can think of.

There is no plausible scenario in which Apple implements a new N3E chip using A17 core designs. None.

 

[Cirillo Gherardo]

So when are we getting an M4 Max MacBook Pro?

 

[EugW]

So when are we getting an M4 Max MacBook Pro?

Rumour has it 2024 Q4 to 2025 Q1

 

[tenthousandthings]

That would be a strange way to write about it, but it's the only reasonable explanation I can think of.

There is no plausible scenario in which Apple implements a new N3E chip using A17 core designs. None.

One possible explanation is A18 went into volume production right away in late 2023, before M4, due to expected high demand for the non-Pro iPhone 16 line.

So they used the old “design” to get it into production quickly, to meet projected inventory requirements for September’s launch. They couldn’t wait for M4.