MP All Models Mac Pro X,Y…WHAT IF…

[ZombiePhysicist]

Just the thing for "cashed up hobbyists" right?

Very potent processor. Apple could have had one of these...

They would have sold them too! Yea, not for joe average, but I mean... Jesus the i9 beats the M2 ultra. We have the AMD 96 core, and this. Apple isnt in the same universe. It's so sad.

 

[jimmy_john]

Intel Announces 288-Core Sierra Forest CPU, 5th-Gen Xeon Arrives December 14

Intel’s new behemoth packs a threaded punch.

www.tomshardware.com

In the mean time apple with its apologists deep up its…myopia.

Hard to imagine anyone with a brain putting a 288 core CPU designed explicitly for cloud compute into a workstation that they actually want people to buy.

 

[ZombiePhysicist]

The M3 reveal was kind of lackluster. Even if they could make an M3 extreme, and I’m not holding my breath, not sure it would be competitive on the graphics front, or cpu front.

Maybe others have some insights I’m missing. Max ram would only go up to 512gb. Not terrible but not great either.

 

[Boil]

No M3 Ultra or (hypothetical/theoretical) M3 Extreme yet...

Apple could be holding back LPDDR5X for those, so we could (in theory) see up to 1TB of RAM with a quad-chip model, which could turn out to be 960GB if an in-line ECC method is used...?

M3 Extreme

• 64-core CPU (48P/16E)
• 160-core GPU (w/Dynamic Cache, hardware mesh shading, hardware ray-tracing)
• 64-core Neural Engine
• 960GB ECC LPDDR5X RAM
• 2TB/s UMA bandwidth

 

[Mac3Duser]

I am still happy with my 7.1
For the moment, I don't need the M3 ultra (given the specifications of the m3 max).
The most important thing is, for me, compatibility with Windows and Linux, and Nvidia cards (for AI, and 3D).
So xeons, even with an old configuration, thanks to pcie lanes, with replaceable cards, are essential.
On the other hand, portable Macs are really getting better and better (except for the price of storage).

 

[deconstruct60]

Maybe others have some insights I’m missing. Max ram would only go up to 512gb. Not terrible but not great either.

512G with no ECC perhaps isn't 'terrible' , but is deeply in the highly dubious zone. Even 256GB with no ECC is pretty dubious. I suspect would hand wave as to how that is 'insanely great' , but it is dubious.

In the event, they were pretty chipper about 128GB being a remarkable amount of RAM , so I wouldn't expect Apple to be hyper focused on matching some 512GB threshold at all. Yes, there are users up on that range. But likely Apple has also counted how many there are using Mac Pros.

The M3 Pro being downsized in transistors is a bit dual edged. It is very suggestive that this chips are expensive. ( Apple going smaller in part to hit budget constraints. It isn't being extra sized for giggles. ). That is a bad, bad , bad sign for an "Extreme". Because the Max is even more expensive. And 4 * even more expensive starts to get into the range of 'almost nobody is going to buy it.

The expensiveness is further indicative that the Max now has two Memory bandwidth configurations 300 and 400 GB/s. Only get the 400 GB/s if do the maximum buy of cores. [ suggestive that 300 one could have binned elements in it. ( one bad memory controller/SLC bypassed and not coupled to a memory package but use the die anyway. )

The 'upside' on that M3 Pro forking off of hyper re-use of the CPU core design of the Max is that if the 'underling' doesn't have to hypershare the top 'half' of the Max die design then the 'overling' (above Max) may not have to either.
Maybe the multiple die solution is a more deliberate designed chiplet. ( not forcing there to be 16 Thunderbolt controllers because shoehorned a monolithic chip into chiplet duties. )

 

[deconstruct60]

No M3 Ultra or (hypothetical/theoretical) M3 Extreme yet...

Pragmatically the M3 Extreme isn't even theoretical. Even if Apple wanted to get a Quad made ... who is going to make it?

AMD and Nvidia GPUs Consume Lion's Share of TSMC's CoWoS Capacity

TSMC's CoWoS capacity shortages will persist, says report.

www.tomshardware.com

AMD, Nvidia and other big package system makers have fully booked TSMC packaging for super sized package. Apple is probably stuck on Info-LSI which has a seriously smaller max size limit around a reticle. ( around two Max size dies).

Apple may have made some Frankenstein CoWoS packages for the lab , but at this point there are significant factors there outside their control.


As for the M3 Ultra. Apple relatively just shipped the M2 Ultra. Put those machines on the obsolete/vitange countdown clock in under 9-10 months? Probably not. The folks that did buy them spent a lot. Technically paying more doesn't necessarily get longer support from Apple, but it is a contributing factor as to why new Mac Pro's roll out so slowly.

Apple could be holding back LPDDR5X for those, so we could (in theory) see up to 1TB of RAM with a quad-chip model,

M3 goes through a number of new features to lean a bit less on the memory bandwidth ( why Apple is dropping memory channels on some options without paying a huge performance cost. ). Decent chance Apple spent more time working on those innovations than in building into all M3's a memory controller to run memory there were never going to attach.

If the Max had LPDDR5X that would be a decent indicator. That it doesn't and is also leaning on 'better caching' is suggestive that there is no M3 plan for LPDDR5X. N3B dies are still going to be expensive. Pretty good chance Apple is coupling higher RAM end user costs with even higher margins to somewhat offset fully functional M3 Max die costs. The Ultra version probably can't get away with dropping a memory controller cluster as well as a single Max can. LPDDR5X is going to cost more and which would leave less to offset more expensive dies.


As for 1TB of RAM. Without ECC, that is even more in looney toon land.

Apple's mulitple die solutions need to be more "designed to be chiplets" rather than pouding 'round peg' monolithic dies into 'square hole' multi-die contexts. Don't necessarily need to do 4 compute dies either. More flexibitily but not crazy expensive , Frankenstein project either.

 

[ZombiePhysicist]

Yea I forgot about ECC. Really is necessary at these sizes IMO.

 

[Boil]

With the massive volume of chips Apple orders from TSMC, I am sure they could work a deal for X amount of CoWoS for the Mn Extreme Mac Pro...?

As for 1TB of RAM. Without ECC, that is even more in looney toon land.

Yea I forgot about ECC. Really is necessary at these sizes IMO.

Which is why I mentioned in-line ECC, with a maximum of 960GB LPDDR5X; like Nvidia is using for Hopper Grace...?

 

[deconstruct60]

With the massive volume of chips Apple orders from TSMC, I am sure they could work a deal for X amount of CoWoS for the Mn Extreme Mac Pro...?

work a deal for no slots available ? not likely at all. Apple is a big 'die' customer. They are not a big CoWoS customer. It is extremely likely not worth it to TSMC to piss off the folks who have been fully committed to CoWoS all along for several years over some 'johny come lately'. It isn't like Nvidia , AMD , and other AI folks are 'broke' . They have piles of money too.

Apple didn't buy their way into N3 coming out sooner did they. About the same issue.

Which is why I mentioned in-line ECC, with a maximum of 960GB LPDDR5X; like Nvidia is using for Hopper Grace...?

Apple chopping down memory controllers is going in the OPPOSITE direction of throwing more memory bandwidth at ECC overhead consumption. To do in-line ECC you need an substantive ECXESS of memory bandwidth to cover the overhead; not less.

More then several things that Apple does is all pointing away from ECC. The "apple is in deep love with ECC" fan club is living in an alternative universe.

 

[ZombiePhysicist]

work a deal for no slots available ? not likely at all. Apple is a big 'die' customer. They are not a big CoWoS customer. It is extremely likely not worth it to TSMC to piss off the folks who have been fully committed to CoWoS all along for several years over some 'johny come lately'. It isn't like Nvidia , AMD , and other AI folks are 'broke' . They have piles of money too.

Apple didn't buy their way into N3 coming out sooner did they. About the same issue.





Apple chopping down memory controllers is going in the OPPOSITE direction of throwing more memory bandwidth at ECC overhead consumption. To do in-line ECC you need an substantive ECXESS of memory bandwidth to cover the overhead; not less.

More then several things that Apple does is all pointing away from ECC. The "apple is in deep love with ECC" fan club is living in an alternative universe.

Any chance they cut that bandwidth off for ECC? Meaning, they can bring those channels back for an ECC version of the chip?

 

[Harry Haller]

What if you could get a Mac Pro with the new Threadripper HEDT and Pro CPUs.
16-96 cores, 1-2TB ram and the latest Quadra Ada GPUs.
Imagine al the 3d, motion graphic, game dev, AI and arch viz folks coming back to Apple.

 

[deconstruct60]

What if you could get a Mac Pro with the new Threadripper HEDT and Pro CPUs.
16-96 cores, 1-2TB ram and the latest Quadra Ada GPUs.

macOS tops out at 64 threads. over 32 cores with SMT or 64 without buys what?

 

[jimmy_john]

What if you could get a Mac Pro with the new Threadripper HEDT and Pro CPUs.
16-96 cores, 1-2TB ram and the latest Quadra Ada GPUs.
Imagine al the 3d, motion graphic, game dev, AI and arch viz folks coming back to Apple.

Yeah I mean all Apple would have to do is change its entire corporate strategy of vertically integrated hardware and software, drop Metal for CUDA, and rewrite major portions of its operating system.

 

[mattspace]

Yeah I mean all Apple would have to do is change its entire corporate strategy of vertically integrated hardware and software, drop Metal for CUDA, and rewrite major portions of its operating system.

When Apple was at its previous peak of vertical integration and proprietary single-platform graphics drivers, is when they nearly went broke.

The entire recovery from that time, was a story of ditching in-house for industry standards, and becoming an included player in a wider tech ecosystem.

That's the history.

 

[jimmy_john]

When Apple was at its previous peak of vertical integration and proprietary single-platform graphics drivers, is when they nearly went broke.

The entire recovery from that time, was a story of ditching in-house for industry standards, and becoming an included player in a wider tech ecosystem.

That's the history.

Well you're not wrong but you're missing the greater context.

All of that was true, but back then, they actually needed the Mac to survive as a company, and Mac Pro was a big part of that.

That is obviously not the case now. The Mac Pro is a vanity project at this point, from a financial perspective.

 

[mattspace]

Well you're not wrong but you're missing the greater context.

All of that was true, but back then, they actually needed the Mac to survive as a company, and Mac Pro was a big part of that.

The greater context is that every company which tried to build a business on proprietary operating system, AND proprietary / low volume processor has gone broke in the effort.

SGI went broke doing it (notwithstanding their death throws with Intel NT machines).
SUN went broke doing it.
Commodore went broke doing it.
Digital went broke (and were eaten by HP because they were non-viable) doing it.
Be went broke doing it.
Apple went broke doing it.

Once a company or platform gets into the mindset of "well we'll only serve a subset of users, and anyone who wants what we don't do can go somewhere else" the only option they have is to squeeze their diminishing returns. The growth story ends there, and you become a rentseeker - which is literally Tim Cook's business plan.

 

[jimmy_john]

The greater context is that every company which tried to build a business on proprietary operating system, AND proprietary / low volume processor has gone broke in the effort.

SGI went broke doing it (notwithstanding their death throws with Intel NT machines).
SUN went broke doing it.
Commodore went broke doing it.
Digital went broke (and were eaten by HP because they were non-viable) doing it.
Be went broke doing it.
Apple went broke doing it.

Once a company or platform gets into the mindset of "well we'll only serve a subset of users, and anyone who wants what we don't do can go somewhere else" the only option they have is to squeeze their diminishing returns. The growth story ends there, and you become a rentseeker - which is literally Tim Cook's business plan.

If Apple's only revenue source was workstations, your historical examples would be apt.

If SGI had diversified their portfolio to include a market dominating mobile phone that accounted for half their revenue, and multiple other lines that accounted for another 40+ % of revenue, relegating their proprietary workstation solution to 5-10% of revenue, well then they'd probably still be around.

 

[mattspace]

If Apple's only revenue source was workstations, your historical examples would be apt.

If SGI had diversified their portfolio to include a market dominating mobile phone that accounted for half their revenue, and multiple other lines that accounted for another 40+ % of revenue, relegating their proprietary workstation solution to 5-10% of revenue, well then they'd probably still be around.

The iPhone is definitely their biggest product, but all the reasons people buy iPhones - the lock-in to other Apple technologies, are being undone by the E.U. The iPhone is primarily a dongle for iMessage, and breaking iMessage exclusivity is guaranteed to be on the E.U's agenda.

The Apple Watch, which was supposed to be the next BIG platform for Apple keeps suffering basic failures of its UI paradigm, because Apple isn't actually very good at UI - something that people have failed to notice, because Apple had so much inherited wealth of previous-generation UI development. Like people thinking Trump is a good businessman, despite his handling of an inherited fortune being worse than the rate of return of a savings account.

The Apple Watch likewise is on a hollow foundation, because it's turning out most of its big ticket technologies and functions are stolen IP (it's about to be banned from import to the US because of its stolen Blood Oxygen reading tech), so the future viability of that is questionable.

The iPad - it's more like a Microsoft Surface now, than it's like the iPad Apple first launched. It's a Surface that is less good at the things a Surface does well (being a computer with a desktop OS). Its advantages - power efficiency & battery life are commoditised characteristics.

Apple Vision? It's an iPad you wear on your face - it's underpowered for the thing people will actually tolerate wearing a headset to do (immersive reality simulation), and it's overpriced for the things it can actually do, given its obsolescence trajectory. People are talking about it being a prototype for "normal glasses that do AR" but that isn't going to happen in a decade-scale timeframe. Companies like Humane will get display-free immersive computing happening for the equivalent tasks, before Apple gets "normal glasses" displays happening.

SGI owned CRAY in the end, diversifying into more expensive niches didn't save them.

The nicest version of standardised solutions is a business where you can surf the achievements of the entire industry, like an index fund tracks the entire economy. 1997 - 2012, that was Apple's business.

Just watch - the primary purchase motivator for the Mac is going to become "to be compatible with a supported macOS" - basically, a rentseeking protection racket for security updates, where newer computers are bought, not to do anything the old computers couldn't do (in many cases they do less than the old machines), but merely bcause they have a more up to date security pantomime.

 

[jimmy_john]

The iPhone is definitely their biggest product, but all the reasons people buy iPhones - the lock-in to other Apple technologies, are being undone by the E.U. The iPhone is primarily a dongle for iMessage, and breaking iMessage exclusivity is guaranteed to be on the E.U's agenda.

The Apple Watch, which was supposed to be the next BIG platform for Apple keeps suffering basic failures of its UI paradigm, because Apple isn't actually very good at UI - something that people have failed to notice, because Apple had so much inherited wealth of previous-generation UI development. Like people thinking Trump is a good businessman, despite his handling of an inherited fortune being worse than the rate of return of a savings account.

The Apple Watch likewise is on a hollow foundation, because it's turning out most of its big ticket technologies and functions are stolen IP (it's about to be banned from import to the US because of its stolen Blood Oxygen reading tech), so the future viability of that is questionable.

The iPad - it's more like a Microsoft Surface now, than it's like the iPad Apple first launched. It's a Surface that is less good at the things a Surface does well (being a computer with a desktop OS). Its advantages - power efficiency & battery life are commoditised characteristics.

Apple Vision? It's an iPad you wear on your face - it's underpowered for the thing people will actually tolerate wearing a headset to do (immersive reality simulation), and it's overpriced for the things it can actually do, given its obsolescence trajectory. People are talking about it being a prototype for "normal glasses that do AR" but that isn't going to happen in a decade-scale timeframe. Companies like Humane will get display-free immersive computing happening for the equivalent tasks, before Apple gets "normal glasses" displays happening.

SGI owned CRAY in the end, diversifying into more expensive niches didn't save them.

The nicest version of standardised solutions is a business where you can surf the achievements of the entire industry, like an index fund tracks the entire economy. 1997 - 2012, that was Apple's business.

Just watch - the primary purchase motivator for the Mac is going to become "to be compatible with a supported macOS" - basically, a rentseeking protection racket for security updates, where newer computers are bought, not to do anything the old computers couldn't do (in many cases they do less than the old machines), but merely bcause they have a more up to date security pantomime.

You should try the Vision Pro some day. The idea that it’s just an iPad strapped to your face is pretty comical.

 

[mattspace]

You should try the Vision Pro some day. The idea that it’s just an iPad strapped to your face is pretty comical.

It's more like an iPad than it's like a Mac - it's more like an iPad than its like any other headset-based system. The vast majority of its apps will be iPad apps with a minimum-effort port to VisionOS - just like Catalyst apps are garbage secondrate apps on Mac, and iOS App on Mac even moreso.

It's hard to get an app that actually takes advantage of Mac specific features now (not even Apple makes them any more), the idea that an even lower volume, more expensive device is going to be different to that is somewhat comical to me.

 

[jimmy_john]

It's more like an iPad than it's like a Mac - it's more like an iPad than its like any other headset-based system. The vast majority of its apps will be iPad apps with a minimum-effort port to VisionOS - just like Catalyst apps are garbage secondrate apps on Mac, and iOS App on Mac even moreso.

It's hard to get an app that actually takes advantage of Mac specific features now (not even Apple makes them any more), the idea that an even lower volume, more expensive device is going to be different to that is somewhat comical to me.

It didn’t take long for me to want one after trying it. It’s really quite remarkable. Even if you only use it as a Mac monitor it is a great value for what it is, relative to a $4000 XDR.

 

[ZombiePhysicist]

So despite having 25% less bandwidth the initial numbers look like m3Max is equivalent of m2Ultra.

M3 Max benchmarks show M2 Ultra performance in a MacBook

Initial Geekbench scores for the M3 Max place the 16-inch MacBook Pro on par with the Apple Silicon Mac Pro.

appleinsider.com

Kind of makes their cadence annoying. You barely get the M2ultra out only to have a laptop chip smoke your 'Mac Pro'. Lame.

 

[mattspace]

M3 Max benchmarks show M2 Ultra performance in a MacBook

Initial Geekbench scores for the M3 Max place the 16-inch MacBook Pro on par with the Apple Silicon Mac Pro.

appleinsider.com

Kind of makes their cadence annoying. You barely get the M2ultra out only to have a laptop chip smoke your 'Mac Pro'. Lame.

The GPU numbers will be my interest - comparing what AUD$12k YoY gets you for an Ultra spec Mac, compared with AUD$3-4k for the biggest consumer Nvidia, or AMD Pro GPUs.

Dual 4k, 5k, 8k 3D performance, etc.

 

[Mac3Duser]

So despite having 25% less bandwidth the initial numbers look like m3Max is equivalent of m2Ultra.

M3 Max benchmarks show M2 Ultra performance in a MacBook

Initial Geekbench scores for the M3 Max place the 16-inch MacBook Pro on par with the Apple Silicon Mac Pro.

appleinsider.com

Kind of makes their cadence annoying. You barely get the M2ultra out only to have a laptop chip smoke your 'Mac Pro'. Lame.

I don't understand the apple strategy about "macs"