MP All Models 40 core ASI with PCI-E slot is around the corner!!

[prefuse07]

Obviously you're just spitballing here, but why would it be desirable for the machine to have an Apple x86 chip? I'd have thought there would be a lots of good reasons for wanting such a machine to use an Intel or AMD processor:

- ASi's 'secret sauce' would provide no advantage in this context
- Apple are unlikely to be better at making x86 CPUs than Intel / AMD
- A handful of Mac Pro sales wouldn't support competitive x86 CPU development
- The MP would always have parity with the best of the PC market
- It would make continued x86 MP releases much more likely
- CPU upgrades would be easy and cheap
- Windows compatibility is guaranteed
- Etc. etc.

I'm not disputing the general concept though - I can see the appeal.

Only because then it's still "Apple Silicon", and they can say "The transition to Apple Silicon is now complete. From now on, we will make two chip variants, an ARM variant which is light on power usage, for all of our machines up to Studio, and an x86 chip for the Mac Pro only, which will have a limited production, because we realized that our Pro users don't give a **** about power consumption the same way that a laptop user does, they need complete modularity and upgradeability, and the ability to use legacy x86 apps. And we understand that none of this is possible within the current facets of ARM architecture, but we are working on a solution to that, and you're gonna love it."

This would also be possible because TSMC fabricates the x86 chips for AMD, so they have them on hand.

Of course -- this dream scenario would never happen.

I do agree with all of your points, and on that note, does anyone know if Apple publicly announced any sort of "end of partnership" with Intel and/or AMD?

I know they've been hinting at it, but have they officially said anything? Does anyone know?

Update: This is interesting

https://www.macworld.com/article/676011/amd-continues-to-work-with-apple-despite-arm.html

This is also interesting, but this makes me lose hope in future apple products

https://www.nytimes.com/2022/07/12/technology/apple-jony-ive-end-agreement.html

 

[mode11]

Only because then it's still "Apple Silicon", and they can say "The transition to Apple Silicon is now complete. From now on, we will make two chip variants, an ARM variant which is light on power usage, for all of our machines up to Studio, and an x86 chip for the Mac Pro only, which will have a limited production, because we realized that our Pro users don't give a **** about power consumption the same way that a laptop user does, they need complete modularity and upgradeability, and the ability to use legacy x86 apps. And we understand that none of this is possible within the current facets of ARM architecture, but we are working on a solution to that, and you're gonna love it."

In that case, it would be far easier to just make an ASi chip that supports lots of PCIe lanes. There is no barrier to this with the ARM ISA - e.g. the Ampere Altra supports 128 lanes of PCIe 4.0.

The only reason Apple's designs haven't to date is because a laptop (and by extension, mini, iMac or Studio) doesn't need them. If Apple would be prepared to go to the expense of making a unique x86 design for a few thousand workstations, they could save themselves a lot of trouble and expense by just making a custom version of their own design.

 

[mode11]

Update: This is interesting
https://www.macworld.com/article/676011/amd-continues-to-work-with-apple-despite-arm.html

Could be something, or could just mean AMD provide Apple with some GPU design expertise? Though that's cool in itself.

This is also interesting, but this makes me lose hope in future apple products
https://www.nytimes.com/2022/07/12/technology/apple-jony-ive-end-agreement.html

Can't get past the paywall there, but it's not surprising Apple and Ive have parted ways. That always sounded like bullsh!t really - lots of companies do this kind of 'we look forward to future collaborations' thing, when likely neither side has any intention. It's just spin so investors aren't spooked, with the 'collaboration' quietly shelved later when no one's paying attention.

 

[goMac]

The AMD/Apple continued partnership article is fairly ancient. AMD hasn't commented on working with Apple in a long time - and has actually started issuing benchmarks against Apple chips recently. I wouldn't put too much faith in that old report.

 

[treehuggerpro]

I think Ternus and Borchers both chose their words carefully, to avoid committing Apple to any particular course of action. Whatever Apple's plans at the time the statements were made, it was always possible that market conditions or manufacturing issues would necessitate a change. If you don't make promises, you don't risk breaking them, nor have to explain any failures. The AirPower mat will have reminded them of that.

At the Studio launch in March 2022, John Ternus said "...making our transition nearly complete, with just one more product to go - Mac Pro. That, is for another day." (https://www.youtube.com/live/CUwg_JoNHpo?feature=share&t=3340). This statement wasn't to stoke interest in the Mac Pro, but rather to make clear the Studio wasn't intended to be its replacement (to head off any concern about lack of PCIe etc.). It acknowledged the x86 Mac Pro hadn't transitioned yet, but made no reference to what form its replacement might take (blade server? Cloud service? Traditional box? Mac Studio Pro? etc.).

In March 2023, Bob Borchers could easily have said 'from MacBook Air to Mac Pro', but he didn't. Whilst the ASi version is still unreleased, the Mac Pro is a long-standing (and current) part of the Mac line up, so it shouldn't have been controversial to mention it in a broad-brush context. If Apple were e.g. discussing iOS, they obviously wouldn't provide the specs of the next iPad, but it's not a secret there will be one. It was weird that Apple were (and still are) being so coy about the very existence of the future Mac Pro, theoretically mere months from its unveiling. Again, this may suggest that whatever solution Apple has for high-powered Mac workflows may take a different form than a traditional workstation.

"Transitioning fully to Apple Silicon" and taking "our entire product line to Apple Silicon" only implies there will be no x86 left in the line up - not that there will be an ASi equivalent of every previous x86 model. If Apple had e.g. replaced the entire Mac range with an ASi laptop, it would have still met that objective.

Complex conclusions.


Yes, neither of their statements have foreshadowed what the Mac Pro would be or when it would be delivered. I wasn’t speculating on these.

Basically, I read both of their (typically restrained) statements as straight forward indications that the transition remains incomplete, that is, until they have delivered an Apple Silicon alternative for their last remaining Intel / Pro Tiered machine. Thus far, it appears typical Apple to me, minimal comment and slow, or late, on delivery.

The only thing I would speculate from the fact that we have had statements on the Mac Pro to date, is that, for whatever reason, their original timeline has slipped enough that they felt it warranted indicating (twice) that the ASi Mac Pro was still in the works.

 

[mode11]

their original timeline has slipped enough that they felt it warranted indicating (twice) that the ASi Mac Pro was still in the works.

Fair enough, though if the intention was to remind us that an ASi Mac Pro is in the works, why say "we believe strongly that Apple Silicon can power and transform experiences from the MacBook Air to all the way up to the Mac Studio"? Why not openly say 'up to the Mac Pro'? Mentioning it would reveal nothing about its configuration or release date.

It was also a statement about the potential of Apple Silicon - "we believe". It wasn't a comment about the products they already sell, e.g. "Apple Silicon is doing a great job, from the MacBook Air to the Mac Studio".

Or did he just misspeak, and I'm reading too much into it (quite possible)?

 

[mattspace]

This is also interesting, but this makes me lose hope in future apple products

https://www.nytimes.com/2022/07/12/technology/apple-jony-ive-end-agreement.html

IMHO Ive's value to Apple (as opposed to value to Apple Share Price) is overstated. I think there's a serious revisionist examination of his legacy due, because Ive at Apple before Jobs was largely middling, and his post-Jobs Apple work has been almost universally garbage - notches, dysfunctional thinness / smallness, unworkable UI. Post-Apple there's been nothing of note from him.

Everything that he did that is good, is stuff that Jobs was directing, which says a lot, again IMHO.

Also, Teenage Engineering is where the true successor to Dieter Rams' (mixed with the best of 1980s Sony) design language can be found (if only it weren't so expensive).

 

[leman]

Apple just released an updated version of their 2020 patents where they describe an advanced cost-efficient 2.5D packaging technology for placing two dies on one package. The scope of this new patent seems be be much more focused compared to the previous more general one. Apple often publishes this kind of patents just before new technology hitting the market (happened with M1, M2, and more recently M2 Pro/Max). Doesn’t have to mean anything but could also be that a high-density superchip is about to make its debut.

P.S. There is also a fresh RT patent which condensed the patents published last year.

 

[leman]

Only because then it's still "Apple Silicon", and they can say "The transition to Apple Silicon is now complete. From now on, we will make two chip variants, an ARM variant which is light on power usage, for all of our machines up to Studio, and an x86 chip for the Mac Pro only, which will have a limited production, because we realized that our Pro users don't give a **** about power consumption the same way that a laptop user does, they need complete modularity and upgradeability, and the ability to use legacy x86 apps. And we understand that none of this is possible within the current facets of ARM architecture, but we are working on a solution to that, and you're gonna love it."

This would also be possible because TSMC fabricates the x86 chips for AMD, so they have them on hand.

Of course -- this dream scenario would never happen.

This would kill the Mac. How is that a dream scenario?

 

[mattspace]

This would kill the Mac. How is that a dream scenario?

How would it kill the Mac? The processor is irrelevant to macOS, except insofar as having the same processor as the rest of the industry was the best thing to ever happen to it, and the rest of its history with unique processors has been slow deflation.

 

[majus]

Gurman's much-ado-about-nothing "newsletter" today is about as interesting as picking belly-button lint out of our navels.

 

[steve123]

Apple just released an updated version of their 2020 patents where they describe an advanced cost-efficient 2.5D packaging technology for placing two dies on one package.

Do you have a patent / publication number(s)?

 

[Boil]

Gurman's much-ado-about-nothing "newsletter" today is about as interesting as picking belly-button lint out of our navels.

No need to slander the Bard like that, tying Gurman's drivel to Will's work...! ;^p

 

[dgdosen]

Apple just released an updated version of their 2020 patents where they describe an advanced cost-efficient 2.5D packaging technology for placing two dies on one package. The scope of this new patent seems be be much more focused compared to the previous more general one. Apple often publishes this kind of patents just before new technology hitting the market (happened with M1, M2, and more recently M2 Pro/Max). Doesn’t have to mean anything but could also be that a high-density superchip is about to make its debut.

P.S. There is also a fresh RT patent which condensed the patents published last year.

Could this be related? a 2x Ultra would be large... https://www.anandtech.com/show/1887...ize-super-carrier-interposer-for-extreme-sips

 

[leman]

Do you have a patent / publication number(s)?

It's this one: https://patentscope.wipo.int/search/en/detail.jsf?docId=US398111768&_cid=P11-LI7S3U-16093-1

 

[leman]

Could this be related? a 2x Ultra would be large... https://www.anandtech.com/show/1887...ize-super-carrier-interposer-for-extreme-sips

Who knows. I suppose a knowledgeable person could read between the lines. I'm just an intertested amateur, sadly...

 

[leman]

How would it kill the Mac? The processor is irrelevant to macOS, except insofar as having the same processor as the rest of the industry was the best thing to ever happen to it, and the rest of its history with unique processors has been slow deflation.

Because Apple just spend years developing a new hardware infrastructure with unique features and APIs, and convincing the industry to use them. And now one is talking about doing a 180 degree turn and asking the industry to use support two different paradigms. The unique value of Apple Silicon is that it is different, and that you have to embrace the difference. The moment you dilute this, you lose the value. Macs are a whole experience, not just a generic PC brand. Apple sells a system, not a box.

 

[steve123]

It's this one: https://patentscope.wipo.int/search/en/detail.jsf?docId=US398111768&_cid=P11-LI7S3U-16093-1

Thanks, that is interesting. I could not find anything that looked unique to a four die implementation but there is enough there that I think it reasonably describes the Ultra packaging.

i was looking more closely at the SEM images of the TechInsights M1 Ultra teardown and it appears to be different than TSMC InFO-LSI in several important but subtle ways. First, it appears like they form a distinct "processor package" that consists of the two M1 Max die placed on a wafer level package (WLP) carrier chip first, the UltraFusion silicon interposer is bonded to the Max die and an organic interposer is formed with a couple wiring layers. The processor packages are then singulated into individual "Ultra Chiplet" and bumped. Second, they form a distinct intermediate interposer in WLP form with TIV and embedded passives and bond a singulated Ultra Chiplet to the interposer chip last. The second WLP wafer is singulated. Then a singulated assembly from the second WLP wafer is combined with the memory chips using a PoP package assembly method with the package substrate. The patent you referenced appears to cover all these aspects.

 

[steve123]

Could this be related? a 2x Ultra would be large... https://www.anandtech.com/show/1887...ize-super-carrier-interposer-for-extreme-sips

They would not need something quite that large for a four Max die package. Now that I think I understand the packaging steps they can build the "processor package" and the second interposer using existing WLP fan out techniques and reticle limits. The final step is PoP with the memory chips but I do not think that step has restrictive reticle limits. They key piece is the UF interposer ... how to design that to connect all four die.

 

[mattspace]

Because Apple just spend years developing a new hardware infrastructure

It's just a variation of ARM

with unique features and APIs,

Everything it can do, can be done on competing hardware, and its early power / performance advantages have largely vanished.

and convincing the industry to use them.

which "industry"?

And now one is talking about doing a 180 degree turn and asking the industry to use support two different paradigms. The unique value of Apple Silicon is that it is different, and that you have to embrace the difference. The moment you dilute this, you lose the value. Macs are a whole experience, not just a generic PC brand. Apple sells a system, not a box.

The processor is not a paradigm. Onboard graphics with no upgradability is not a paradigm, these are just limitations. This is just a way to replace the Mac with an iPad running macOS.

The "value" of the mac, was that it was a better operating system, and whenever its difference from the general PC market was greatest, so was it's collapse in popularity and industry relevance.

People who don't remember the 68k, and PowerPC days are doomed to repeat them.

 

[deconstruct60]

It's this one: https://patentscope.wipo.int/search/en/detail.jsf?docId=US398111768&_cid=P11-LI7S3U-16093-1

That looks much more like what Apple would do to put a discrete cellular modem ( with digital subsystem part ) in an A-series package than something that would be applied to a Mac Pro class solution .
While Apple has in-housed the modem silicon delevelopment , it is making less and less and less sense to put the modem on the same fab process that the CPU/GPU/NPU cores are on . TSMC N3 forks away from being aligned with what modems need. It just gets worse at N2. Lots of folks have presumed that Apple bought the modem tech so they could slap it all onto the same die as the Arm stuff. That is not nessary at all . If simply just 2 dies in same package with ‘2D’ or ‘2.5D’ packaging that is better Perf/watt than a separate package 10-30mm away.

The majority of the rest of phones that Apple competes with have monolithic dies ( e.g., Qualcomm ) , that made sense when I/O , SRAM , and compute logic were all getting the same increases on each fab elation step . They have all forked off into different directions. Enough so that Apple’s discrete isn’t so bad ( more expensive , but Apple can just make customers pay more ) . But they don’t need to pay quite so high a package to package power loss .

The real challenge is just how high a run rate they can run with this packaging .

The Mac is going to need better than 2.5D packaging. ( and doesn’t have iPhone high run rates )

 

[steve123]

That looks much more like what Apple would do to put a discrete cellular modem

I took a look at that patent application and it captures the aspects of the M1 Ultra packaging as shown in the SEM cross sectional images TechInsights published after they did their teardown. TSMC is assembling the packages but it appears to be a custom package that consists of two integrated discrete packages: a fan-in "processor package", a fan-out package substrate with PoP bonded memory. Evidently Apple has patented this package. The same patent could also be used for aspects of a package the way you describe as well. There was nothing in the patent application that I saw that revealed anything that you could say was definitely "quad" related. Though again, this general patent could in fact cover a quad die assembly as well.

The thing about the Ultra package is the UF bridge only has to connect two die. In a quad die arrangement, to maximize performance and minimize latency each die likely has to be connected to each other die (the die are not "bussed"). This implies the number of interconnections is 6x the number of connections between the two die in the Ultra. The increased number of connections would increase the size of the UF bridge. Do we know how many connections UF uses between the two die in the Ultra package?

 

[leman]

Thanks, that is interesting. I could not find anything that looked unique to a four die implementation but there is enough there that I think it reasonably describes the Ultra packaging.

i was looking more closely at the SEM images of the TechInsights M1 Ultra teardown and it appears to be different than TSMC InFO-LSI in several important but subtle ways. First, it appears like they form a distinct "processor package" that consists of the two M1 Max die placed on a wafer level package (WLP) carrier chip first, the UltraFusion silicon interposer is bonded to the Max die and an organic interposer is formed with a couple wiring layers. The processor packages are then singulated into individual "Ultra Chiplet" and bumped. Second, they form a distinct intermediate interposer in WLP form with TIV and embedded passives and bond a singulated Ultra Chiplet to the interposer chip last. The second WLP wafer is singulated. Then a singulated assembly from the second WLP wafer is combined with the memory chips using a PoP package assembly method with the package substrate. The patent you referenced appears to cover all these aspects.

Are you suggesting that this patent might be describing M1 Ultra instead? I was under impression that M1 Ultra didn't use SoC die stacking. The patent describing the M1 Ultra and UltraFusion should be this one: https://patentscope.wipo.int/search/en/detail.jsf?docId=US295683435&_fid=WO2020112505

Besides, the "high-bandwidth die to die interconnect" patent explicitly talks about building an SoC as an 3D arrangement of dies. This is not a technology Apple is currently shipping. Some quotes from the patent: "in particular, such folded die arrangements may be used to split SoC cores into separate dies", "the first die 140 may be a main chip including higher performance cores (e.g. CPU, GPU) or cores fabricated with smaller node technology, while the second die 110 may be a daughter chip including lower performance cores (e.g. RF, memory) or cores fabricated with a larger node technology"

That looks much more like what Apple would do to put a discrete cellular modem ( with digital subsystem part ) in an A-series package than something that would be applied to a Mac Pro class solution.

Or it could be used to solve their problem of increasing manufacturing costs on cutting edge nodes. They could put compute clusters on 3nm while keeping cache/display controller/memory controllers on 5nm. This application is heavily hinted in the patent.

But you are right that I initially mistook it for a way to connect two symmetric dies (like two Max-chips). Apple is explicitly talking about splitting SoC functionality across dies.

 

[leman]

It's just a variation of ARM

It's not just a "variation of ARM", it's a scalable unified compute architecture that requires substantial software overhaul to be used effectively. Especially if we talk about the pro software.

Pushing for Apple use x86 for the MP is like asking Nvidia to drop CUDA on their high-end GPUs and use AMD's HIP instead, or asking Sony to use DX12 on half of their playstation models. Might make some sense to users who only care about their short-lived computing needs (I want this particular software version to run now!), doesn't make any sense for the company or the future of the platform, and is utterly disastrous for long-term business planning.

which "industry"?

The pro software industry obviously. Which else?

 

[mattspace]

It's not just a "variation of ARM", it's a scalable unified compute architecture that requires substantial software overhaul to be used effectively. Especially if we talk about the pro software.

No, its *just* a variation of ARM, that reinvents shared graphics memory, which was described by Apple as "vampire graphics memory" when they weren't using it. Just because an attractive hairstyle in a polo-neck shirt says it's "innovative" on a recorded slide deck, and a cottage industry of press-release-regurgitators cosplaying as journalists repeat it, doesn't mean it is.

As for it being "scalable", in comparison to what? IA-64 x86-64 (typo) goes from embedded NAS processors to Workstations and Severs. "Minimally scalable" is the most accurate definition for Apple Silicon at the moment - high performance cellphone, to low performance computer is not the great breadth of scale you may think.

Pushing for Apple use x86 for the MP is like asking Nvidia to drop CUDA on their high-end GPUs and use AMD's HIP instead,

Except that X86 offers higher absolute performance at the top end, lower power draw for performance on mobile, and wider hardware & software compatibility.

Gods, it sounds like an absolute nightmare to which one is sentenced .

or asking Sony to use DX12 on half of their playstation models. Might make some sense to users who only care about their short-lived computing needs (I want this particular software version to run now!), doesn't make any sense for the company or the future of the platform, and is utterly disastrous for long-term business planning.

Why do you think the fate of the Mac will be any different on Apple Silicon, to that which it experienced when it was on m68k, and PowerPC. It's still a minority platform, that's now back to having to have an entire dedicated codebase (that never matures because Apple are always changing things, to keep developers "on their toes").

Apple's problem during the x86 era was never that their computer processors weren't fast enough, and it was never that they drained their batteries too quickly. Apple's problem was consistently that they designed bad cooling solutions, that they included weak graphics, and made inefficient, low-performance & buggy software.

None of that is going to change as Apple removes the ability to directly compare their products with their competitors'.

The pro software industry obviously. Which else?

Which pro software industry? Graphics, Film & Video, Gamedev, anything to do with content creation, engineering, architecture, 3D, visualisation... none of these industries are tied to macOS or Apple Silicon. None of them were crying out for a new processor architecture to support.