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

[prefuse07]

or Elon purchased apple

That idiot couldn't afford it, no matter how leveraged he is

 

[Mago]

That idiot couldn't afford it, no matter how leveraged he is

That idiot soon will be on mars by literally owns mean, meanwhile NASA is spending as much money as Elon on whole Starlink constellation just on sending 4 dudes to the moon, no recyclable neither transplanetary capabilities unless an huge revision on the whole SLS/Orion not yet even funded and given NASA hyper inflated cost (actually a elite community subsidy system) likely to cost 4x or more than whole Starlink (as figure consider the JWST cost as much as a nuclear super carrier or two Virginia class subs).
Elon is fair to insulting him cause hes an BIG moron, but a moron that delivers. On the other side we have polite people just burning our money with utter shamelessness.

And being serious, Elon won't Buy apple he can build his own Apple and quickly outsell Apple just investing a small fraction on what would cost buying stocks to control apple.

 

[prefuse07]

 

[majus]

...he can build his own Apple...

And just who would be clamoring to buy an Elon Pro without a proven track record? Not me.

 

[Mago]

This is exactly backwards if N3B is 'very , very bad' for high volume. The plain Mn die is an over 10M/year run rate die. ( The MBA , MBP 13" , iMac 24" are over 50% of Mac sales . At about 20M/yr that 10M right there. Still haven't included the iPad Pro and iPad Air. ).

Pretty good chance that all the 'drama' about N3B having really bad yields is pretty overblown at this point. N3B extremely likely costs substantially more (perhaps 'bad' in an environment where most vendors are looking to cut unit costs to protect margins) , but Apple is pretty brave at throwing additional component costs at user pocketbooks rather than their own.

N3B : scalable ( 2 or more )
N3E : M3 , M3 Pro , laptop only Max

might be true. If Apple started all of them on N3B design and then later did a substantive redesign for N3E , then the scalable one was already in the expense zone. Throwing even more design overhead costs at it only would make it even more expensive. The price is already high and the run rates relatively low so is far , far better position to absorb incrementally lower yields. (Apple's mark up is large total amount on these high end chips per unit). The 'scalable' model is not a high volume SoC. It is much bigger , but not high volume. ( yields are already incrementally lower just being bigger). If the denser cache memory of N3B was the dominant source of the yield problem that would push all of them over to N3E. There would be a problem if the disproportionately higher cache memory presence was a root cause problem issue. But that isn't a 'high volume' issue. N3B yields are probably incrementally lower than N3E yields but Apple is charging 100's of dollars more for the 'scalable' chips also. The discarded dies are being paid for.




Going through a substantive redesign for N3E would incur a substantively delay. M2 just came out in 2022. It isn't even a year old. Sliding into late 2023 / early 2024 isn't any worse than the M1 -> M2 time gap. M2 Pro/Max are even younger!!! Apple needs new ones of those 'soon' like they need another hole in the head. ( The Pro isn't a "hand me down" SoC. Once replaced in a MBP 14/16" or Mini Pro it is gone from the product line up. Max is relatively "fall off a cliff" worse. )


That said. I'd be surprised if Apple 'split up' the M-series across two different design nodes. If N3B was so 'bad' financially it chased away A17/M3 then pretty good chance the rest will follow suit. And the 'scalable' would take a price and time hit ( e.g., the very low top end variant disappear because now 'too expensive' by saddled with a 'designed twice' cost. ) . Apple would take negative hit on Mac Pro 'timely arrival' though.

All on N3B in 2023 and wait for N3P for some limited line updates next year also would be tractable (presuming the doom and gloom about yield are overblown. ) N3B takes longer to fab, but Apple can work around timing with a good plan and temporarily willing to carry some incrementally higher SoC inventory than usual periodically.


The only Apple SoC that has a better chance of being completely split off is the A17. There are lots of 'hand me down' products for that so the economies of scale are just way better than even the M3 (let alone the rest of the M3 era line up). But timing wise it would be rather risk if Apple wants the iPhone Pro models to come out on time. N3E probably won't be in HVM production with enough lead time to hit a demand bubble at the end of September.

N3B probably will be a worse fab process for any 'long term' product that is going to be used in multiple generations of Apple products. If Apple wants to play 'hand me down' if the A17 into a plain iPad in a couple of years then probably worth taking a short term rollout out hit, to have a SoC that can make for several years.

Mostly I agree with your analysis, only differ on i consider the MBP13 will never see another generation (you may agree with me), and mba13/15 don't outsell MBP14/16 , the iMac likely to move close to the Mac mini having M3/M3pro variant, over this landscape has sense to offer n3b for MBA+non pro iMac, and reserve N3E for high transistor high volume soc as M3pro Max and scalable (we may have no M3 max but only the scalable version).

Apple betting on m3 scalable on N3E, still possible but how to fit the upcoming MBA13/15.

I consider less risky build M3 on n3b as it implies far less transistor indeed favoring a new node given yields specially decrease as transitor count rises, an smaller SOC less affected, meanwhile the M3 scalable unlikely to be as small as to discard it yield impact, add this scalable CPU likely to have higher clocks and operate on rude workloads, conservatively you may agree prudent to skip n3b for scalable apple silicon.

Consider each scalable ASi SOC to include 4+16 cores as rumoured and at least twice GPU cores even More than M2 Max, that's a lot of silicon.

 

[deconstruct60]

post-twitter Elon can barely afford to buy AN apple.

Even having pissed 20-30B down the drain Musk isn't broke. That said some kind of 'levered buyout' of Apple ... even two or three large entities would have trouble doing that in tighter credit markets. Apple isn't some deeply depressed , undervalued stock. If anything it is too high.


Musk pissing around on stunts like taking the 'w' out of Twitter to put a 'Titter' sigh on HQ ... Say what you will of Cook , he doesn't spend much time on childish pranks and half-baked tweets composed while stoned. There is very little evidence he couldn't handle the complexity of running Apple at all. Without a relatively large talent pool , Apple can't be Apple. ( Apple isn't really a 'one man show'. It wasn't ever just Jobs. Nor is it now just Cook. )


Musk took many millions from other folks to assist in buying Twitter. He is pretty much of track to pissing away many of their millions too. Good luck getting any money out of them the next time he wants to do some 'mind fart' acquisition. Musk's likely bigger loss is likely going to be repetitional over the long term. He is probably never going to be a non millionaire , but getting other folks money (other than from his cult) is going to be harder.

 

[Mago]

And just who would be clamoring to buy an Elon Pro without a proven track record? Not me.

Elon FYI now has what apple failed: it's own machine learning cloud on their own silicon (TPU).

Lets dream, assume Elon buys Ampere or Fujitsu (easy on their reach), and develop a family of servers/workstation/PC based on HCC RISC-V on an well money managed friendly Linux distro one that doesn't rely only on community to quickly solve issues and provide support, given an standard controlled (while theoretically open) hardware is provided quality and user experience are easier to address, this is exactly what Linux desktop has been waiting: solid reliable predictable hardware where surprises are discarded and money flow to keeps The software well maintained and friendly.

FYI migrating/coding to Linux Now easier than ever ever for some services/applications even easier than windows, and no macOS is as stable or performant as an solid Linux distribution, where you don't have proprietary nightmares as planned obsolescence or hardware segregation (as breaking with Nvidia).

I think this day is coming with or without Elon, and apple don't looks well at this landscape.

 

[Mago]

Even having pissed 20-30B down the drain Musk isn't broke. That said some kind of 'levered buyout' of Apple ... even two or three large entities would have trouble doing that in tighter credit markets. Apple isn't some deeply depressed , undervalued stock. If anything it is too high.


Musk pissing around on stunts like taking the 'w' out of Twitter to put a 'Titter' sigh on HQ ... Say what you will of Cook , he doesn't spend much time on childish pranks and half-baked tweets composed while stoned. There is very little evidence he couldn't handle the complexity of running Apple at all. Without a relatively large talent pool , Apple can't be Apple. ( Apple isn't really a 'one man show'. It wasn't ever just Jobs. Nor is it now just Cook. )


Musk took many millions from other folks to assist in buying Twitter. He is pretty much of track to pissing away many of their millions too. Good luck getting any money out of them the next time he wants to do some 'mind fart' acquisition. Musk's likely bigger loss is likely going to be repetitional over the long term. He is probably never going to be a non millionaire , but getting other folks money (other than from his cult) is going to be harder.

Controversial but as business operation Twitter is now way better than a year ago when was close on being broke, traffic an engagement skyrocket while cost plummeted, painful for some Yes, the brightest business operation not, but is working, and that Elon stunt as remove the W from Twitter banner is the way he tells how in control he is and the business is going without consequences.

 

[dgdosen]

So the announcement is some live stream?

What's worse than watching YouTube click harvester content? Watching YouTube click harvester content they're too lazy to edit.

Thank you for linking up - that'll make it easier to ignore content from the lot.

 

[prefuse07]

So the announcement is some live stream?

What's worse than watching YouTube click harvester content? Watching YouTube click harvester content they're too lazy to edit.

Thank you for linking up - that'll make it easier to ignore content from the lot.

Yeah no thanks, I'll keep waiting for some news to post here on or one of the other tech news sites. Screw these youtube clowns.

 

[mattspace]

Even having pissed 20-30B down the drain Musk isn't broke. That said some kind of 'levered buyout' of Apple ... even two or three large entities would have trouble doing that in tighter credit markets. Apple isn't some deeply depressed , undervalued stock. If anything it is too high.

30billion down the drain on Twitter, another 30 billion euros potentially to be levied in fines in Germany for Twitter's refusal to remove Nazi content. Tesla stock which he's facing stock manipulation and labour practices lawsuits over, and can't sell for fear of exposing himself to lawsuits for crashing the shareprice. Then his Twitter co-investors, some of whom are the sort of folks who cut people up with bone saws when displeased.

*popcorn*

 

[deconstruct60]

Mostly I agree with your analysis, only differ on i consider the MBP13 will never see another generation (you may agree with me), and mba13/15 don't outsell MBP14/16 ,

You are on drugs. The MBA 13" ( M2 and M1 ) likely outsells both he MBP14 and 16 combined all by itself. Add in a MBA 15 priced in the same slot where the MBP 13" is (the MBP 13" being the second more popular Mac Apple sellls) and those two combine will still sell pretty close to 35+% of all Macs. ( Apple mentioned it explicitly last June MBA #1 seller. MBP 13 #2 seller. )

After Apple released the Mini (well 'half' of the Mini line up) , MBA , MBP 13" , iMac they crossed more than 50% of Mac sold being on M-series.

Apple Now Selling More M1 Macs Than Intel-Based Models, Says Tim Cook

Despite only being released in November, sales of the M1-powered MacBook Air, MacBook Pro, and Mac mini now represent the majority of Mac sales,...

www.macrumors.com

That was significantly before the MBP 14"/16" shipped. MBP 14/16" were not the Macs that solely pushed the Mac M-series mix way over the 50% line.

Apple might squeeze out more aggregate profits from the 14"/16" combination... but unit sales? Probably not.

Apple could screw up the MBA 15" by pricing it incrementally higher than the MBP 13". At that point they would kill the hold the MBP 13" has in the #2 position. But if they do 'kill that position' then it would not be surprising for the MBP 13 to stay around. Either on the M2 ( like the trailing more affordable M1 MBA) or for the small but loyal touchbar fans. But the easier thing to do would be not screw up the pricing ( just MBP 13" like internals in a bigger case with no fan and just incremental more screen+battery costs driven increase. Dropping touchbar screen helps offset increase in main screen size costs increase. )

Apple would be pretty silly not to somewhat control costs of the MBA 15". If look in the mainstream laptop market, there are gobs of decent 15" $600-999 priced models. Yeah they don't have "retina" screens, but there is a large market segment that wants a larger screen, but doesn't want to pay $2K for one. Apple isn't going to match the main market price points but the closer to $999 they get the significantly more they will likely sell.

The M2 MBA 13" is already 13.6" ( if just rounded up that would be 14"). So 14.6 - 14.9 rounded to 15" wouldn't particularly be a very expensive screen cost increase. Nor would it huge cap the 14.2" MBP in screen estate. Just allows user to push more of the component cost toward the screen as opposed to the SoC.

The 14"/16" mainly differentiated on faster , more robust processors for those that need 'horsepower' is probably more than healthy product segmentation value.

the iMac likely to move close to the Mac mini having M3/M3pro variant,

Unless Apple comes up with another enclosure for the iMac ... it likely isn't going to get a M3 Pro variant. Internal space wise Apple painted themselves into a corner on the iMac 24".

So far it doesn't seem like they are all that unhappy with that constraint. Also more than pretty doubtful they want to sell less Studio Displays. So large screen iMac (with associated larger internal space ) seems doubtful also. If Apple deeply wanted a large screen iMac they could have tossed an M1/M2 Pro into a iMac Pro case at any time along the way relatively extremely easily. They didn't.

I suspect the iMac 24" stays exactly like it is. It gets a M3 bump, because the iMac is just going to iterate slower than the rest of the line up. It suffers the same , back-burner, state the Mini used to get when it updated every 4 years as a lower priority project. It is not the Apple anointed primary desktop system anymore.

over this landscape has sense to offer n3b for MBA+non pro iMac, and reserve N3E for high transistor high volume soc as M3pro Max and scalable (we may have no M3 max but only the scalable version).

Same fallacy you started off with before. Max isn't a high volume SoCs. Not even relatively close.
The bulk of MBP 14"/16" sales are quite likely Pro ; not Max. The unit volume of the plain Mn versus Mn Max is likely about an order of magnitude difference.

Apple betting on m3 scalable on N3E, still possible but how to fit the upcoming MBA13/15.

I consider less risky build M3 on n3b as it implies far less transistor indeed favoring a new node given yields specially decrease as transitor count rises, an smaller SOC less affected,

The M2 Max is substantively bigger than the M1 Max and yet Apple did it anyway. To some extent the binned around it. The M2 has 10 GPU cores so a 10 core cluster building block. The M2 Pro maxes out at 18 cores (not 20). and the Max at 38 cores ( not 40).

With N3B you can make that die smaller than the M2 Max size ( on relatively old N5P). So you did decrease the die size (hence pulled back some incremental yield. ). Keep the same incremental binning of the "extra" GPU cores may not necessarily need all 100% of and ta-da .. about slightly smaller die and probably probably about similar yields You don't want the scaling die to be as big as possible. Unless trying to design a bad/worse chiplet.

If have something that is scalable then can 'peel off' some stuff and make the chip smaller and just assemble on the package to 'grow'. For example peel off the 'top' of the Max's I/O stuff. PCI-e , SSD , Thunderbolt, etc. and have a smaller resulting die.

There is a faction of folks who want to go to N3-family process so that Apple can throw an even bigger "everything and the kitchen sink" into the die to make it bigger. Perhaps Apple could be using N3 to make the dies smaller? That is kind of what N3 is 'better at'. Especially relative to multiple year old N5P.

Adding in a targeted Perf/Watt effective hardware RT won't necessarily be a huge transistor count increase. Especially if the GPU core count didn't increase (just got better cores as opposed to more cores).

meanwhile the M3 scalable unlikely to be as small as to discard it yield impact, add this scalable CPU likely to have higher clocks and operate on rude workloads, conservatively you may agree prudent to skip n3b for scalable apple silicon.

The I/O ( and perhaps Display Controllers) aren't exactly diminishingly small elements.

Apple unveils M2 Pro and M2 Max: next-generation chips for next-level workflows

Supercharging MacBook Pro and Mac mini, M2 Pro and M2 Max feature a more powerful CPU and GPU, up to 96GB of unified memory, and power efficiency.

www.apple.com

The 4 Thunderbolt controller element (upper right corner and inward about an inch ) of the die is bigger than the E core complex. Bigger that about the P core building block complex. (about 1/4 of P core allocation) . Take that off and have done some shrinking even before apply N3 (B or E flavor).

There are probably defects in 3D package creation but if doing packaging of mulitple dies anyway... that overhead is coming anyway.

Consider each scalable ASi SOC to include 4+16 cores as rumoured and at least twice GPU cores even More than M2 Max, that's a lot of silicon.

If have a scalable chiplet ... why increase the core count on the individual chiplets. Either the chiplet is scalable ( and can get to increased core counts that way) or it isn't really scalable. ( and would have better 'bang for the buck' to scale the memory+ cache controllers... which are not as great match with N3-family anyway).

N3E cache memory size is EXACTLY the same as N5P that Apple is using for M2. Apple's designs are hugely cache heavy. (a very hefty chunk of the Pro die photo above is L1/L2/L3 cache which won't get any smaller at all on N3E. ) The current caches of M2-family won't get any smaller at all with N3E ( and relatively limited incrementally smaller with N3B) . So trying to substantively increase the core count with the cache size the exactly the same is a relatively dubious move. It probably won't work well for CPU cores. And if chasing very high clock, single threaded drag racing scores probably not much there either.

As I said before though. If the fancy fabrication gymnastics that TSMC used to get get to the smaller N3B cache sizes are a major contributor to the yield problems then smaller, but still very cache heavy M3-family designs aren't really helped all that much. N3B costs more per wafer. So even though still get a decent number of 'good' dies out of the wafer , the wafer also costs more. So those are more expensive 'bad' dies to pay for with an individual die that doesn't have as large of a end user price tag assigned to it. If it is very bad for cache for some of the M3 line up then it is likely bad for all of the line up no matter which die size. The only thing that will cover it up is extra money. (price can charge the end user for the die. ).

 

[deconstruct60]

30billion down the drain on Twitter, another 30 billion euros potentially to be levied in fines in Germany for Twitter's refusal to remove Nazi content. Tesla stock which he's facing stock manipulation and labour practices lawsuits over, and can't sell for fear of exposing himself to lawsuits for crashing the shareprice. Then his Twitter co-investors, some of whom are the sort of folks who cut people up with bone saws when displeased.

*popcorn*

Twitter is still a corporation. 30 Billion in fines leverage against Twitter isn't necessarily Musk's money. Same thing with all the lawsuits against Twitter. Good luck getting blood from a turnip. One reason why Musk doesn't give a flip if all he is doing is driving into a big hole in the ground. It is twitter that is going to get sued and twitter is broke.

Typically, Tesla hasn't made any money and the mostly cultish stockholders hung around. As long Price cuts put a floor under the stock price and can go back to making no profits , he'll probably mainly get a pass.

The $30B stock asset would transfer ownership. That doesn't necessarily mean the new owners would irrationally sell it in a panic. Twitter wasn't going to repay the loan that quickly anyway. So not having it super soon wasn't in the cards either way.

Twitter didn't make money before Musk bought it. The fact that is doesn't make money after he bought also puts his co-investors into the position of "who is the bigger fool? The fool or the fool who follows the fool. ". Why did they think they were going to make money in the first place. The whole thing was structure counting on some even bigger fools to come along and buy it off of them for an irrational price after doing some 'facade restructuring'.


Musk has a "I'm a multi billionaire " security team. Those bone saw folks mainly like picking on folks who can't defend themselves.

 

[Boil]

Too much talk about a narcissistic manchild billionaire arsehat...

Let's discuss what the ASi Mac Pro might be...

• "Standard" Mn Ultra / Mn Extreme SoCs...?
• "Asymmetrical" Mn Ultra / Mn Extreme SoCs; "regular" SoC(s) paired with "GPU-specific" SoC(s)...?
• Nvidia Grace / Hopper SuperChip style...?
• AMD multi-chiplet style...?

Still thinking M2 Ultra Mac Pro for a Spring 2023 Mac Event, otherwise M3 Ultra / M3 Extreme if pushed to WWDC 2023...?

Regular & Asymmetrical options would be best, allowing more GPU cores for those who need them, and "standard" CPU/GPU ratios for those who don't...?

 

[dgdosen]

N3B : scalable ( 2 or more )
N3E : M3 , M3 Pro , laptop only Max

might be true.

Given that N3B is in "volume production" are you thinking Apple makes a WWDC announcement for Mac Pro (and Studio) with 3nm chips? up to 4x'Max'?

Would they use the name M3 (M3 Ultra/M3 Extreme)? That would imply an M2 Air 15" as well.

 

[prefuse07]

That would imply an M2 Air 15" as well.

Yup, looks like you're on the money

 

[Adult80HD]

Too much talk about a narcissistic manchild billionaire arsehat...

Let's discuss what the ASi Mac Pro might be...

• "Standard" Mn Ultra / Mn Extreme SoCs...?
• "Asymmetrical" Mn Ultra / Mn Extreme SoCs; "regular" SoC(s) paired with "GPU-specific" SoC(s)...?
• Nvidia Grace / Hopper SuperChip style...?
• AMD multi-chiplet style...?

Still thinking M2 Ultra Mac Pro for a Spring 2023 Mac Event, otherwise M3 Ultra / M3 Extreme if pushed to WWDC 2023...?

Regular & Asymmetrical options would be best, allowing more GPU cores for those who need them, and "standard" CPU/GPU ratios for those who don't...?

I highly doubt you'll see any "asymmetrical" setups. Seems like a weird choice. I expect standard M3 Ultra/Extreme chips and PCIe support with an Apple GPU/accelerator card for ML/AI. But we shall see soon, no doubt.

 

[Boil]

I highly doubt you'll see any "asymmetrical" setups. Seems like a weird choice. I expect standard M3 Ultra/Extreme chips and PCIe support with an Apple GPU/accelerator card for ML/AI. But we shall see soon, no doubt.

Not a weird choice if one has a need for a CPU/GPU core ratio that favors the latter...

 

[Adult80HD]

Not a weird choice if one has a need for a CPU/GPU core ratio that favors the latter...

Yes, but not a trivial matter in terms of engineering the interconnect/fabric bridge. Not likely to be worth the effort.

 

[leman]

Yes, but not a trivial matter in terms of engineering the interconnect/fabric bridge. Not likely to be worth the effort.

I don't think that one would need to use a special bridge. Look at this images (this is from a recent Apple patent illustthat describes their fabric topology):

As you can see they have a plurality of networks (some specialised for GPU-GPU communication and some for CPU-CPU communication) and network switches. The die-to-die bridge simply connects two on-die networks, essentially making a larger chip. With this approach, I don't see why a GPU-only asymmetrical die would be any different. It could use exactly the same bridge interface as the SoC die, but it's CPU network will be either sealed off or passed though.

There is actually at least one interesting advantage in using asymmetric dies — heterogenous memory interfaces. A GPU-only desktop die could use faster, more power-hungry RAM (such as HBM or even GDDR) to support higher bandwidth requirenments of a GPU. You'd still have unified memory and unified cache with all their benefits, but some memory controllers would support higher bandwidth and they will be located closer to the potential bandwidth consumer. I have no idea whether Apple will go for this type of solution, but it could make for a very compelling compute-oriented product. This kind of scheme is only used in some state-of-the art supercomputer tech — if Apple manages to bring it to a desktop workstation, it could be a game changer.

 

[Mago]

This kind of scheme is only used in some state-of-the art supercomputer tech

Actually being replaced by Apu-like solutions on general purpose super computers, ae:

Nvidia Grace Hopper (Venado-LANL and many others)
AMD MI300 (el Capitan-LLNL and some others)

There also other projects with similar architecture excluding GPU replacing it with TPU,NPU, DPU whatever xPU

 

[Apple Knowledge Navigator]

So what was this "big announcement"?

 

[innerproduct]

So what was this "big announcement"?

It was just us collectively being (willfully) fooled by a bunch of youtube-kids …

 

[Mago]

It was just us collectively being (willfully) fooled by a bunch of youtube-kids …

Yes💩

 

[avro707]

It was just us collectively being (willfully) fooled by a bunch of youtube-kids …

what actually did get released?

Looking on the Apple site could only see a Mac mini at AUD$5000 available to purchase (do folk really need Studio) or a MacBook Pro.

I don’t watch these rumours very much and even less noisy YouTube kids.