You're Apple. You just built a 40-core SoC for Mac Pro. Now what?

[senttoschool]

You're Tim Cook, sitting in his nice office, looking at how much money you just spent to make this giant SoC for a relatively small market. In fact, you have to do this every year or every two years to keep the Mac Pro relevant. How do you recuperate some of this money spent?

I might have nailed this.

The Mac Pro is just using an Ultra SoC. No Extreme SoC.

I still think an Extreme SoC is coming but it will come with a cloud version and a local version.

I think Apple did the math and it made no sense to spend so much engineering resources on a chip that will sell in the thousands - not millions.

 

[Spaceboi Scaphandre]

I think the one thing I'm kinda disappointed in with the new Mac Pro is the fact you can't change out the SoC. The Mac Pro could've become a new platform where you use the one chassis and just swap out the component modules when it comes time to upgrade instead of buying another $7000 Mac Pro, but of course that's not the Apple way dammit.

At least being able to swap out the M2 Ultra for the M3 Ultra when that came would've made up for the lack of RAM expansions.

 

[Longplays]

Mac Pro M2 Ultra with 24-core CPU & 32-core Neural Engine comes in

- 60-core GPU
- 76-core GPU (+$1,000)

I think the M2 Ultra 2-die (Extreme) did not appear due to yield and if there is enough demand for it.

Would a Mac Pro M2 Extreme be saleable at $10k?

It would easily fit a Mac Pro tower.

3nm M3 Ultra will likely be out Q1 2025.

By then still no refresh for the 2020 iMac 27".

 

[leman]

I think the M2 Ultra 2-die (Extreme) did not appear due to yield and if there is enough demand for it.

Yield is fine. Connecting four dies is tricky. M2 doesn't have this technology.

 

[senttoschool]

I think the M2 Ultra 2-die (Extreme) did not appear due to yield and if there is enough demand for it.

As leman said above, it's literally just 4x Max dies. They're manufactured separately so the yields for an M2 Extreme would be identical to one M2 Max.

 

[deconstruct60]

As leman said above, it's literally just 4x Max dies. They're manufactured separately so the yields for an M2 Extreme would be identical to one M2 Max.

Actually what leman just said is that it's literally is NOT just 4x Max dies. The Max die does NOT have the interconnect facility or interupt mapping to handle 4 dies (and appear 'smooth' as just one big die to the apps/os ).

The 4 die model pragmatically would have been a different die building block used on even fewer SoCs. In other words, Apple is using MBP 14"/16" Max users to subsidize getting the Ultras made cheaper by slapping limitations on the Max die.

As long as Apple sticks to the model where the laptop users have to paid for features they can't use to make the "Ultra" work, the Extreme has major problems. The problem is not just more pervasive and capable interconnects between the dies. It is also affordability.

 

[Longplays]

As long as Apple sticks to the model where the laptop users have to paid for features they can't use to make the "Ultra" work, the Extreme has major problems. The problem is not just more pervasive and capable interconnects between the dies. It is also affordability.

Would moving to a 3nm die shrink address that concern?

In my head a MBP 16" would be able to handle a M3 Ultra in terms of physical size, power consumption and thermals because it is a now a 3nm chip.

Intel demonstrated that there is a demand for Ultra-class chips on laptops.

Those buying a M3 Ultra laptop would readily accept shorter battery life in exchange for for a portable Mac Studio.

 

[leman]

In my head a MBP 16" would be able to handle a M3 Ultra in terms of physical size, power consumption and thermals because it is a now a 3nm chip.

Usually new chips are the same size or even buffer than their predecessors, to
Implement new features and extract more performance.

In other words, why would Apple utilize 3nm to make a smaller Ultra when they could make a much faster Max?

Intel demonstrated that there is a demand for Ultra-class chips on laptops.

Hardly. Intel demonstrated that selling power-limited desktop CPUs in a laptop chassis is a good advertising move.

 

[deconstruct60]

I might have nailed this.

The Mac Pro is just using an Ultra SoC. No Extreme SoC.

I still think an Extreme SoC is coming but it will come with a cloud version and a local version.

Placing a silicon die in a remote ( or even down the hall) data center room doesn't make extremely custom , very low volume packages any cheaper. If the Extreme SoC is so expensive few can put a 'use value' on it that will get them to buy one , then distance doesn't matter.

I think you are trying to point at some financial model where the SoC is leased out in fractionals. One group can't afford it but if 4 folks share 1/4 the costs then they'll all be happy. The problem is that macOS isn't licensed that way. Nor does the 64 macOS thread limit help a lot either.

And any push to "cloud version" as being competitive with the general Linux Server ( build a SoC focused not on macOS but on Linux ) market has its own additional costs which will get piled on top of the expenses also. Apple is way behind Neoverse/Graviton/Ampere in that space. Paying lots of extra money to catch up isn't going to be cheap.
[ And Apple has major other silicon to do. R1 is going to need to get followed up with an R2 . etc. Still don't have a deployed cellular modem after well over $1B. etc. ]

XCode Cloud can work very cost effectively (at the price points it charges ) with just "off-the-shelf" macs and some modest add-ons to deploy them. They don't need a SoC just for that.

I think Apple did the math and it made no sense to spend so much engineering resources on a chip that will sell in the thousands - not millions.

IF Apple did a chiplet that could be amortized over all of the Mac Studio (MS) and Mac Pro sales then probably could get close to a million. It would mean cutting the MS loose from "co sharing" on the monolithic dies for laptops R&D. But laptops have volume ( most of Apple's sales). Just slicing off the Mac Studio Ultra sales probably falls just as short as the Mac Pro ones.

If Apple needed another "desktop" model if still short of a million for MS + MP then either bring back "iMac Pro". Or do a "Mac on a Card" ( drop a Max onto a PCI-e card and sell to the hypermodular focused folks who are running away from the new MP paradigm of no 3rd party GPUs and no DIMMS. ) . Or a 'half sized' Mac Pro ( toss most of the slots , shrink the power supply , kill the 'space frame' thing. ... more 'boring' and lower bill of materials. )

Apple's 'silicon' strategy is primarily built on putting the SoC into more than one product to leverage better economies of scale. Just one product isn't going to work. But inventing a market that Apple really ,really doesn't want to be in ( Apple canceled macOS Server a while back. Pretty big clue whether they 'love' that market or not) is likely not going to get a multiple product deployment for the SoC.

In short, Apple just needs a desktop focused building blocks to go into a several desktop models. Do that and there is volume that cracks the 'millions' threshold.

 

[deconstruct60]

Would moving to a 3nm die shrink address that concern?

TSMC N3 wafers cost substantively more , not less. Not really helping with costs or "interconnects" at all. And the Interconnect really are not going to shrink.

N3 is probably going to sizably push the needle toward using chiplets (as opposed as "Max" as possible monolithic dies) as a more economical approach. But Apple's almost monomaniacal pursuit of Pref/Watt keeps the tension high for monolithic laptop dies. This issue is whether Apple easies up just very incrementally on that to do better desktop product coverage ( or not ) .

In my head a MBP 16" would be able to handle a M3 Ultra in terms of physical size, power consumption and thermals because it is a now a 3nm chip.

No. N3 isn't going to make the RAM modules any smaller at all. To be effective with Apple's 'poor man's' HBM implementation you have to keep adding more memory channels and RAM die stacks to keep the higher number of GPU cores feed with enough data to be useful. The cores shrinking don't matter if have left behind the I/O they need to be useful.

Throw on top that the cache/SRAM and other non-RAM I/O off the die stops shrinking with N3 and it is not "the whole entire package will get smaller" situation at all.

N3 will allow Apple to put more 'logic stuff' into the same size dies (and very similar sized packages) , but the overall SoC likely is not getting any smaller. The Max SoC barely fits on MBP 16" logic board ( one reason why it is have to do a 4-die combo due to the placement of the RAM controllers to consume long edges of a narrow-ish rectangle. The shape of the M2 Max package is to limbo and squeeze into the narrow shape available on the laptop logic boards. )

Intel demonstrated that there is a demand for Ultra-class chips on laptops.

Not as thin as the MBP 14"/16".

Those buying a M3 Ultra laptop would readily accept shorter battery life in exchange for for a portable Mac Studio.

Apple doesn't really do "desktop replacement" laptops ( go back to the really old MBP 17" model level of thickness and weight) . They are primarily focused on the space where tossing the dGPU out the window is the primary goal.

 

[gpat]

Would moving to a 3nm die shrink address that concern?

In my head a MBP 16" would be able to handle a M3 Ultra in terms of physical size, power consumption and thermals because it is a now a 3nm chip.

Intel demonstrated that there is a demand for Ultra-class chips on laptops.

Those buying a M3 Ultra laptop would readily accept shorter battery life in exchange for for a portable Mac Studio.

M3 Ultra being made on 3nm would just mean that they would cram more cores inside it.
Thermal envelope would likely remain the same.
You could expect a M3 Max being equal to the M2 Ultra and running cool enough for a MBP, I'll give you that much.

 

[deconstruct60]

As long as Apple sticks to the model where the laptop users have to paid for features they can't use to make the "Ultra" work, the Extreme has major problems. The problem is not just more pervasive and capable interconnects between the dies. It is also affordability.

Would moving to a 3nm die shrink address that concern?

No, not in and of itself. In some aspects it actually gets worse with TSMC N3! N3 wafers are more expensive. This subsidity is leveraged on making laptop folks pay for a section of die that is worthless to them. If the wafer is more expensive then this wasted die is more expensive also.

So for some rough examples. If for 2M Max dies Apple was wasting 100 wafers on effectively useless UltraFusion connector, then for a $14K wafer then comes to a $1.4M 'waste tax' . For $20K N3 wafer then is a $2M 'waste tax'. If N2 goes up to $22K wafer costs , then it is a $2.2M 'waste tax'. As the wafer costs go up ... so does the tax laptop Max users.

It is a bit worse than that because the UltraFusion is not going to shrink. N3 is only a decrease in size of logic ( think of that as circuits that compute). The I/O ( which is a major component of what UF does) and SRAM/cache does not. So the I/O and SRAM parts of the N3 dies are getting more expensive also with N3 wafers (and will again with N2).

It is going to make less and less sense to inflict a chiplet connector on something that is primarily intended to be a monolithic die. The SRAM and I/O not shriking will likely eventually force Apple off the monolithic path in parts of the laptop line up, but I think Apple will 'fight' that move for as long as they possibly can. It is just incrementally better Pref/Watt to stay monolithic and Apple isn't charging 'deep discount' prices for Mn , Mn Pro , or Mn Max SoCs.

In my head a MBP 16" would be able to handle a M3 Ultra in terms of physical size, power consumption and thermals because it is a now a 3nm chip.

You are doing shallow system analysis. The RAM packages are not on N3. So they aren't going to shrink at all. The Ultra is STILL going to need just as many RAM packages as it has now! The only pragmatic way to squeeze an Ultra onto the laptop board space constraints would be to throw away memory bandwidth. That means throwing away performance. So would buy some 'bragging rights' system that ran great "suck benchmark into local die cache and run it a brief time" , but not particularly great broad real world app performance.


The Mn Max dies have monstrous amounts of SRAM L2/L3-"System Level cache". N3 is going to next to nothing for those in terms of area saving; no 'shrink'.

They also have a relatively large number of memory controllers. Where those controllers go out to reach out the RAM packages ... again very diminished gains (or at least very limited ) gains. May get 'fancier , but almost free' interactions with the Memory packages ( some 'on the fly' memory compression , ECC computations , etc. ) but the path off the die isn't going to shrink as much.

Intel demonstrated that there is a demand for Ultra-class chips on laptops.

Intel is doing NOTHING like Ultra-class chips in laptops. Only at the most superficial level are they lined. If spinning about Meteor Lake is using 4 tiles/chiplets and 3D interposers that part is similar. But the I/O throughput is not even in the same zip code , let alone a demonstration.

And what Intel is doing is not more affordable. They are doing this in part because they have to. They don't have the equipment to make enough Intel 4 dies to populate most of the tiles on Meteor lake set up. If AMD chooses to get into a very hefty 'price war' with Intel , then Intel would be boxed into a corner. AMD probably won't because they have a ton of expensive bills to pay also. But they also have the same problem because TSMC probably can't cover all of the Intel 4 dies production either on top of their other commitments. [ AMD just needs to boils the frog slowly to pull share away from Intel over time. ]

Those buying a M3 Ultra laptop would readily accept shorter battery life in exchange for for a portable Mac Studio.

It isn't just as shorter battery. It would be a substantially thicker case also. And with max dense aluminum option taken that is substantially more weight.

Apple could get some more logic board space to place the additional Memory packages by just going to a twice as big single fan ( big as in taller. ). But that bigger single fan likely will more often have to spin faster and generation substantively more noise.

So battery life ... diminished. Lightweight ... diminished . Low noise ... diminished.

The final major problem is that Apple only has a limited amount of Industrial Design throughput. They are not built to do 10 different Mac models like Dell/HP/Lenovo are for Windows. Those folks try to make everything for everybody. Apple does not. The new MBA 15" very likely puts the MBP 13" at threat of being dropped. I suspect Apple won't drop it for M3, but it would lumber on in a comatose state as far as the Industrial design went. If the MBA 15" does better than expected then it probably dies.

The number of Mac models expanded a bit with the MBA 15" but that is only because another model is going comatose on design updates.

Apple does the MBP 16" so they don't have to do a 17-18" models. Another "bigger and luggable" laptop ? there isn't tons of unit volume in that space to support it. Ditto why Apple isn't doing a xMac box with slots in between the Mini and Mac Pro.


When Apple gets to a Mn Max on N3P or N2 that will be close enough spitting distance to the M1/M2 Ultra that it won't matter. Apple has resources to 'afford' the time to wait for the tech to come to them, rather than chase it now into a niche of a niche upper end laptop market with yet another form factor they don't have structure to chase after.

 

[dmccloud]

The biggest reason you will most likely not ever see an M2 or M3 Ultra in a notebook form factor is readily apparent in the teardowns of the M1 Ultra Mac Studio. The heatsink is massive to the point where it accounts for most of the thickness differences between the Studio and Mac Mini. While the move to M2 and possibly 3nm with an M3 variant can lower power consumption to a small degree, the SoC would still be generating large amounts of heat. Consequently, there still needs to be a beefy cooling solution to vent heat away from the SoC.

 

[deconstruct60]

M3 Ultra being made on 3nm would just mean that they would cram more cores inside it.

N3 doesn't mean you can get more Cache on the same size die. More cores without more cache doesn't necessarily buy a tremendous leap.

Pretty good chance see better cores (with more internal logic resources) , more so than significant uptick in cores that have large cache requirements attached to them. Going from two 4-P cores cluster + one 4-E core cluster to still two 4-P core cluster to two 4-E core cluster. Likewise incremental bumps on GPU cores and NPU cores and more robust Video en/de coding (cover AV1 but no more additionals to concurrent streams coverage). Similar on displays. M3 unsurprisingly still sporting the same, now even more lame, 2 stream output, but maybe now finally DisplayPort v2.1 support.

Thermal envelope would likely remain the same.

If Apple isn't adding the maximum number of performance cores possible then could see some thermal decreases. The overriding problem though is that the overall Ultra SoC package footprint probably will not shrink at all. N3 does nothing for I/O so the connections to off-die other parts of the package are not going to shrink much at all. The fanout is just as big to the other components.

The thermals will drop a bit, but not limbo down into 'easy to cover in relatively thin laptop' land. The job of the Studio's cooling system will probably get better ( less strained over most workloads. Whether that is low enough to dump the copper material may/may not happen. )

You could expect a M3 Max being equal to the M2 Ultra and running cool enough for a MBP, I'll give you that much.

it isn't going to run that cool because not even going to have the same number of memory packages. Let alone N3 doesn't promise anywhere near that kind of savings.

N5 vs N3
" Power savings -25-30% ..."

TSMC Details 3nm Process Technology: Full Node Scaling for 2H22 Volume Production

www.anandtech.com

There is no 50% drop there at all! And that 20-30 only comes if toss the performance uptick completely out the window. Can go max faster or get minimum, lower power, but trying to do both comes with trade-offs.

The likelihood of Apple completely throwing performance out window to claim the 30% crown is pretty dismal. Apple is already behind the x86_64 laptop market in single thread. They are going to need to lean on some of that N3 performance uplift just to stay attached to the market forces. Throwing all of that away would just mean falling even further behind. And even worse situation on the desktop placements for the Ultra.

Apple is going to throw more work at E-cores so in contexts where just sitting at a table looking at the Macrumors webpage and just reading ( and running some widgets to sprinkle information the screen) the N3 SoCs will do a bigger job. But 'foot on the gas' doing very active professional workloads, pretty good chance not going to see any huge battery life gains at all. (some but not huge).


P.S. Yet another reason to fork the 'desktop' , always multiple die base building block die off from the laptops. two 4-P + two 4-E makes some sense for the laptop market , but if going to have multiple dies present all the time, that extra 4-E doesn't make lots of sense. Another something-else that was small would make more sense (GPU cores or NPUs ) . Already going to get another 4-E cluster when pair up the dies. At 8 E-cores just how much backgroud stuff going to scrouge up. An incrementally smaller die might even help also if going with chiplets ( makes them more cost effective to make. Like two UltraFusion connectors would be better so could segment off the lost of the non-memory I/O into a cheaper to make chiplet. )

 

[Longplays]

The biggest reason you will most likely not ever see an M2 or M3 Ultra in a notebook form factor is readily apparent in the teardowns of the M1 Ultra Mac Studio. The heatsink is massive to the point where it accounts for most of the thickness differences between the Studio and Mac Mini. While the move to M2 and possibly 3nm with an M3 variant can lower power consumption to a small degree, the SoC would still be generating large amounts of heat. Consequently, there still needs to be a beefy cooling solution to vent heat away from the SoC.

The active HSF is that robust to comply with silent PC & silent room requirements. I believe there is a MR thread about Mac Studio owners whining about coil whine and how "noisy" their Mac Studio is.

There is a market for workstation laptops and the users are more than willing to put up with short battery life, heat and noise. They likely need it portable enough to travel with and they'll plug it into the wall over 80% of the time.

If an M2 Ultra, that is two M2 Max chips "fused" together, uses 2x more power then it means that the MBP 16" battery life would be cut 1/2. That is still longer than most workstation laptops.

Seeming Ultra chips consumes far less than 240W than an Intel Core i9-13980HX laptop with 330W charger then there must be a market for it.

When Apple has a surplus on Ultra chips then I see it happening. In the same way that Ultra chips only appeared in the 2023 Mac Pro because Apple couldn't supply enough of them for the 2021 Mac Studio.

It may appear as early as 15 months from now with 2024 Mac Studio M3 Ultra & 2024 Mac Pro M3 Ultra.

But if Apple wants to make them compliant with silent PC & silent room requirements all they need to do is 2x the thickness to 3.1cm and 2x the weight to 3.26kg. It would likely sell for $5.5k-6.5k.

 

[redheeler]

Mark Gurman is saying that Apple is working on a 40-core SoC for the Mac Pro for 2022.

You're Tim Cook, sitting in his nice office, looking at how much money you just spent to make this giant SoC for a relatively small market. In fact, you have to do this every year or every two years to keep the Mac Pro relevant. How do you recuperate some of this money spent?

You create "Apple Cloud". No, not iCloud. Apple Cloud. Like AWS. Where anyone can come and rent a 40-core M3 SoC running on macCloudOS. You get into the cloud hosting business. You file this under the "Services" strategy that you keep pushing to make Wall Street happy.

Soon, you'll be releasing 64-core SoCs with 128-core GPUs, then 128-core SoCs with 256-core GPUs, and so on. Somehow, you're actually beating anything AWS, Azure, Google Cloud can offer... without really trying.

Apple Silicon Cloud.

It wouldn't surprise me if Apple is already testing their own SoCs to power their iCloud service, which currently depend on AWS. Apple was reportedly spending $30m/month on AWS in 2019. It might be $100m+ per month by now given how fast services have grown.

Building cloud infrastructure isn’t something that can be done easily if you’ve a shiny new SOC. You need global distributed data centers. You need network infrastructure. You need software for load balancing. AWS, Azure, Google Cloud, etc. have all this and many years of experience maintaining it.

The web has come a long way from the times where a single data center full of powerful servers could suffice for the speed and reliability of a service like iCloud. Today the bottlenecks are better resolved with improved network infrastructure, caching and load balancing algorithms, and closer geographic location of the servers more than anything.

 

[jlc1978]

There is a market for workstation laptops and the users are more than willing to put up with short battery life, heat and noise. They likely need it portable enough to travel with and they'll plug it into the wall over 80% of the time.

The question is how profitable is that market? If the margins aren’t there and having it isn’t necessary, ignore it.

 

[Longplays]

The question is how profitable is that market? If the margins aren’t there and having it isn’t necessary, ignore it.

Great question... that product segment could be a loss leader and halo product to Intel given how niche it is.

Ship at most 25,000 annually? A proton to an atom in a bucket of water that is Apple's revenue?

Maybe they're worth as much as Mac Pro users demanding upgradeable SoC, memory, storage, dGPU and other parts?

Let us see by 2H 2024 for M3 Ultra to appear for a MBP 16", 17" or 18" with it.

Going back to topic if Apple managed to make a M2 Extreme non-binned it would have this spec

Chips	M2 Extreme
CPU	48-Core
High-performance	32x
High-efficiency	16x
GPU	152-Core
Neural Engine	64-Core
Transistors	268 billion
Max unified memory	384GB
Memory bandwidth	1.6TB/s

That Mac Pro would likely be $11k or more with a CPU max power consumption of 600W on a 1.3kW PSU.

 

[deconstruct60]

Great question... that product segment could be a loss leader and halo product to Intel given how niche it is.

Apple does not 'do' loss leader product systems. Folks keep trying to spend the money in the Scrooge McDuck money pit... Apple's rich they could do a product that just overtly looses money and "make it rain" on a subset of users. Apple is rich in large resason because they do not do that. They sell product systems that make money, not loose money.

Products where folks don't what to pay for what it costs to make and develop, Apple skips those customers. Apple is not trying to sell everything to everybody. They sell a handful subset of products in major market areas to folks who can pay for them. Product A doesn't have to pay for Product B.

Apple does overtly practice placing major components in multiple product systems to drive up economies of scale savings on making those components. But that is to create LARGER profit margins; not to burn extra money on a money loosing product.

The 24K Watch was aligned with this halo concept and it got dropped by Apple later. It is NOT how Apple product marketing nominally works. Find the folks who want to buy the product actually going to sell to them, not some product they are not going to buy.

 

[Longplays]

Apple does not 'do' loss leader product systems. Folks keep trying to spend the money in the Scrooge McDuck money pit... Apple's rich they could do a product that just overtly looses money and "make it rain" on a subset of users. Apple is rich in large resason because they do not do that. They sell product systems that make money, not loose money.

Products where folks don't what to pay for what it costs to make and develop, Apple skips those customers. Apple is not trying to sell everything to everybody. They sell a handful subset of products in major market areas to folks who can pay for them. Product A doesn't have to pay for Product B.

Apple does overtly practice placing major components in multiple product systems to drive up economies of scale savings on making those components. But that is to create LARGER profit margins; not to burn extra money on a money loosing product.

The 24K Watch was aligned with this halo concept and it got dropped by Apple later. It is NOT how Apple product marketing nominally works. Find the folks who want to buy the product actually going to sell to them, not some product they are not going to buy.

But but but Apple must make all Macs upgradable...

Apple must continue selling Intel Macs and add Apple Silicon with it...

Apple must sell the Mac Pro at $2999...

I demand 1.5TB of RAM!

Not doing this causes e-waste even when eBay, garage sales, and places will buy your Mac

 

[deconstruct60]

But but but Apple must make all Macs upgradable...

The early computers in the 50's had vacuum tubes. If a tube burnt out you could just replace and and move on. All that decoupling and deintergation had trade-offs. Replaceable but also bigger and more power consuming.

TSMC/Intel/Samsung as making N3 , N2 , N1.8 , N1.6 fab processes for smaller and smaller transistors. How going persue that path and not pull more stuff onto a single die? Go from 50Billion transistors to 150Billion transistors, then something is going to get pulled into the 'Black hole' of the main SoC package. It is the path the whole industry has been on for last 80 years. It is slowing a bit , but has not stopped at all.

Apple must continue selling Intel Macs and add Apple Silicon with it...

LOL. When someone gets up and explicitly says they are getting off the bus ... they are probably going to get off the bus.

Even if Apple wanted to stay in x86_64 land (which they don't) ... they'd need to get on AMD bandwagon. AMD is at best 'luke warm' about Thunderbolt and Intel still hasn't throughly proven they have cleaned up their 'mind fart' mess internally. And in the server spaces AMD/Intel adjustements for Arm are chasing after Ampere Computing and Arm Neoverse implementations on Linux ; not the direction that Apple is going.

Apple must sell the Mac Pro at $2999...

Apple only selling a fixed subset of products will result in a constant stream of grumbling and assertions that the sky is going to fall in if Apple doesn't receive their money.

However, Apple is going to get lots of barking here. HP lists the Z4 as their best selling workstation. So there are a sizable number of folks in that price zone, which the $7K price point is very substantively far away from.

And Apple is a bit less frantic about absolutely lowest levels of fratricide. ( M1 Mini and M1 iMac ... Apple gets paid for the M1 either way. ) . There could be another "desktop Max-like" SoC as another option around the Studio. Probably closer to a $3,999-$4,2999 price zone but that is much lower friction point ( ~$1K ) than a $4K one. Probably would have to chop the slots , ports , power supply, and size.

When Thunderbolt 5 comes and if it gets traction there is pretty good chance Apple is just going to hand wave at that instead of a another whole system for the $3,999 price point.

The $2-2.5K price point is over for some alternative to the Mac Studio. The Studio without the slots , power supply , etc is already in that zone. "more stuff" and cheaper.... probably not.

I demand 1.5TB of RAM!

Without ECC , clamoring for > 1TB RAM is a waste of time. The Mac Pro more seriously needs to cover the 192GB range they have now with ECC and make it solidly work before Apple even considers options for moving capacity up.

The 'good news' is that the max RAM capacity for the Max versions of the laptops is creeping up where it should have ECC also ( the rule of thumb is > 128 , it should be on. ). At some point Apple is going to need some solution even outside the desktop space. M3 , M4 , M5 ... somewhere in that sequence they'll need to figure something out.
( 96 -> 128 is only 33% more. That is going to come with newer LPDDR5/6 implementations eventually.) And Apple's markup on RAM is so high; not putting some 'value add' on it is going to skate out onto thinner and thinner ice over time.

 

[senttoschool]

Building cloud infrastructure isn’t something that can be done easily if you’ve a shiny new SOC. You need global distributed data centers. You need network infrastructure. You need software for load balancing. AWS, Azure, Google Cloud, etc. have all this and many years of experience maintaining it.

Apple has their own data centers in Asia, US, and Europe.

 

[deconstruct60]

Apple has their own data centers in Asia, US, and Europe.

That really doesn't make the some huge 'prime time' player though. Apple isn't 'small' but they are not huge (as 'huge' defined these days). Apple does both. Hosts some workload and farms substantive fraction out ( where 'secure data at rest' and other security issues are not important ).

It is vastly run on Linux and nothing really stopping Apple from deploying Ampere Computer (Arm) nodes just like many of the hyperscalers that are much bigger than Apple do. The Ampere One SoC starts off at 132 cores. ( 4x as many arm cores). If trying to consolidate workload of 50-100 million people want to attack that with nodes that are 24 cores at time or 192 cores at a time???? The latter is probably going to do it at lower cost ( $ and energy).

Even if Apple had a 48 core solution, it is still bringing a knife to a gun fight. They don't have something competitive.

For XCode Cloud what Apple has is fine ( and on macOS 'home turf' so extremely competitive in context). However, that is a narrow corner case of Apple's cloud services.

 

[Joe The Dragon]

That really doesn't make the some huge 'prime time' player though. Apple isn't 'small' but they are not huge (as 'huge' defined these days). Apple does both. Hosts some workload and farms substantive fraction out ( where 'secure data at rest' and other security issues are not important ).

It is vastly run on Linux and nothing really stopping Apple from deploying Ampere Computer (Arm) nodes just like many of the hyperscalers that are much bigger than Apple do. The Ampere One SoC starts off at 132 cores. ( 4x as many arm cores). If trying to consolidate workload of 50-100 million people want to attack that with nodes that are 24 cores at time or 192 cores at a time???? The latter is probably going to do it at lower cost ( $ and energy).

Even if Apple had a 48 core solution, it is still bringing a knife to a gun fight. They don't have something competitive.

For XCode Cloud what Apple has is fine ( and on macOS 'home turf' so extremely competitive in context). However, that is a narrow corner case of Apple's cloud services.

and apple will need to things like have ipmi in services and hot swap storage.
Also more then 192GB ram per cpu.

 

[deconstruct60]

and apple will need to things like have ipmi in services and hot swap storage.
Also more then 192GB ram per cpu.

For XCode Cloud? Not really.