Like it or not, Apple needs Mac Pro.

[IvyKing]

Nothing to do with "bit errors" (Mass storage devices/filesystems/drive controllers have been using checksums/parity etc. since long before journaling filesystems) and everything to do with mass storage being non-volatile and - unlike RAM - expected to keep its data structures intact in the face of software crashes, disconnections, power-failures etc. that might happen in the middle of a complex update.

Since few home computers are attached to a UPS, the ability to recover from a power failure is a necessity.

I strongly suspect that reduction in DRAM cell size will make memory bit flips even more common as there isn't much that can be done about muon background. As such, some form of ECC will be required.

 

[jakey rolling]

...because nobody is actually coming up with any solid evidence to show that this is a problem that exists. If you want to claim that all consumer devices need ECC to solve this "risk" then it's your burden of proof to show that the risk exists.

The funny thing is, this should be something that is fairly easy to prove. If ECC provides the "early warning" that Basic75 is saying it does, it would have to exist as a log entry either in the Windows Event Log (or Mac or Linux equivalent) or in the EFI firmware or BIOS in the system. It should be easy for a "home" workstation-class system user to provide actual numbers showing the frequency of ECC-detectable data corruption events.

 

[mcnallym]

Again, even they dont worth it just like Apple. What just because Apple is the only one developing and using their own chips?

Actually HP, Lenovo, Dell you mentioned ALSO make servers along with workstations. Do you see the point now? Both workstations and servers benefits from each other from development after all.

YES it is just because they make there own SoC.

HP Lenovo etc don’t make their own CPU instead they buy in Intel or AMD, where the R&D for developing that CPU is spread across a massively greater number of CPU. Hence why is possible for them to do the updates on the workstations but not worth to Apple. The cost to update to next Xeon is negligible compared to Apple developing a SoC that used for a niche product of Apple customers.

So much of that Xeon development also covered on the Server side which Apple don’t sell eiether.

Apple on the other hand have to spread the R&D for the new SoC across a much smaller number of SoC, so updating the Studio or Pro with a new Extreme SoC is way more expensive and just not effective to do every year.

You actually HAD to ask that!

 

[mcnallym]

We are talking about Apple, not OpenAI. Apple is using their own closed-ecosystem with their own hardware and software. That's a huge difference. Yeah, they could use other services but eventually, they would develop and use their own. Besides, tell that to Apple Intelligence.

Apple also predominantly use Linux Servers on non Apple Hardware.
They may sell an ecosystem to the customer base but that doesn’t mean everything they use internally is Apple badged.

They also use Azure and AWS despite having Apple Cloud Services and they aren’t going to completely eliminate the use of other cloud services.

Apple have quite often dropped products, they used to do there own

wireless boxes
servsers
SAN
time capsule
Laser printers
Injet printers.

They even stopped doing monitors for a while.

Apple clearly do not feel the need to do everything internally so why do they need to do the AI learning on their own hardware and specifically why would they need a Mac Pro for that.

Plenty of other people are developing AI without developing and releasing a workstation. Wether it by simply buying up Nvidia Hardware or like Google and developing own hardware. Google’s hardware nothing like a Mac Pro because as I stated several posts ago you would not start with a Mac Pro as your base for AI hardware, so the fact that have skills in house for a workstation like machine does not mean that have skills for AI hardware.

Why is it that Apple are Unable to do any of this and have to develop a workstation to be able to do AI.

Your opening post was specifically that Apple need the Mac Pro to do this but so far nothing posted showing why Apple need the Mac Pro for this.

Yes have been apparently putting M2 Ultra Studio’s into Racks for some AI work well if can use the Ultra SoC then can simply use the Studio can you not. At which point don’t need the Pro do they.

 

[1d1otic]

Apple also predominantly use Linux Servers on non Apple Hardware.
They may sell an ecosystem to the customer base but that doesn’t mean everything they use internally is Apple badged.

They also use Azure and AWS despite having Apple Cloud Services and they aren’t going to completely eliminate the use of other cloud services.

Apple have quite often dropped products, they used to do there own

wireless boxes
servsers
SAN
time capsule
Laser printers
Injet printers.

They even stopped doing monitors for a while.

Apple clearly do not feel the need to do everything internally so why do they need to do the AI learning on their own hardware and specifically why would they need a Mac Pro for that.

Plenty of other people are developing AI without developing and releasing a workstation. Wether it by simply buying up Nvidia Hardware or like Google and developing own hardware. Google’s hardware nothing like a Mac Pro because as I stated several posts ago you would not start with a Mac Pro as your base for AI hardware, so the fact that have skills in house for a workstation like machine does not mean that have skills for AI hardware.

Why is it that Apple are Unable to do any of this and have to develop a workstation to be able to do AI.

Your opening post was specifically that Apple need the Mac Pro to do this but so far nothing posted showing why Apple need the Mac Pro for this.

Yes have been apparently putting M2 Ultra Studio’s into Racks for some AI work well if can use the Ultra SoC then can simply use the Studio can you not. At which point don’t need the Pro do they.

M2 Ultra is too slow to use cause Apple has no choice but to use their own chips other than Nvidia. They are not even close to TPU servers. And I already told you, workstations are related to servers due to how they work in terms of part. Why do you keep justifying poor hardware for? If you think Mac Studio or Ultra chips are enough, then you are short sighted and will only kill Apple's future.

Clearly, Apple is too limited with hardware resources.

 

[Basic75]

A paper from 2009:

We find that DRAM error behavior in the field differs in many key aspects from commonly held assumptions. For example, we observe DRAM error rates that are orders of magnitude higher than previously reported, with 25,000-70,000 errors per billion device hours per Mb and more than 8% of DIMMs affected by errors per year. [...] Consistently across all platforms, errors occur at a significant rate, with a fleet-wide average of nearly 4,000 errors per DIMM per year.

Which is presumably still relevant, a newer paper from 2018:

More recent DRAM cell fabrication technologies (as indicated by chip density) show higher failure rates

As for the requested anecdotal evidence:

It should be easy for a "home" workstation-class system user to provide actual numbers showing the frequency of ECC-detectable data corruption events.

A friend has 192GB in his (not very busy) home server. Normally he sees about one correctible ECC error every two months, but has also replaced a couple of memory modules because ECC showed them to be(come) flaky.

 

[maflynn]

Your opening post was specifically that Apple need the Mac Pro to do this but so far nothing posted showing why Apple need the Mac Pro for this.

Agreed, but then the OP thinks that workstations are servers which clearly shows his level of ignorance on the topic. I understand we all lack knowledge on many topics, and I don't pretend to be an expert on AI, but when he starts off his thesis with such a mistaken notion, its hard to continue having a discussion

All workstations are server based

As you mentioned, Apple doesn't eat its own dog food when it comes to their infrastructure, mostly, because they don't have have any. They killed their server line years ago.

Regarding the Macpro as a possible AI machine - so far there's been no reasons stated why this could happen. The Mac Pro offers no advantages over a Mac Studio. Add in the fact that Macs only accounting for 8% of the revenue. It comes as to no surprise why we've seen news reports that apple is largely shelving any updates to the Macpro, there's really no ROI on this product.

 

[theluggage]

Regarding the Macpro as a possible AI machine - so far there's been no reasons stated why this could happen.

The other issue here is the increasing concerns (found all over the press) that AI companies are currently over-valued in comparison to any possible future profits, and that the AI bubble is currently fit to burst. Companies like NVIDIA and Meta are "riding the tiger" - they're already all-in on AI - and have no choice but to keep doubling down, but Apple still has the choice to maybe stay on the sidelines until the "re-adjustment" hits.

For better or worse, Apple were late to the game on AI services and AI development - they'd have their work cut out playing catchup with NVIDIA et. al. even if the current growth rate continues, but sinking investment into developing AI training hardware now runs a real risk of going up in smoke when the crash comes. NVIDIA, Meta etc. might survive, but these crashes tend to be last-in, first-out... Even if someone wanted to get a stake in the AI boom right now, I don't think it's the right time to go for the "plucky outsider" rather than someone who actually has product right now.

Apple's decision to license tech from existing AI giants, use third-party training hardware etc. rather than roll their own could prove to be very, very astute.

What Apple do have is a great platform for consuming AI services on iDevices and MacBooks with better power/performance than comparable devices. Better still, these are devices that people will still want when the petals fall off the AI tulip.

M2 Ultra is too slow to use cause Apple has no choice but to use their own chips other than Nvidia.

Nobody is claiming that the M2 Ultra Mac Pro is a good machine for AI training. It's not even for that - it's mainly for users who beed specialist PCIe cards that aren't GPUs/TPUs. There's already a M3 ultra and no reason not to think that there won't be a M4/M5 Ultra in due course. (There may be some cases where having 512GB of unified RAM directly accessible by CPU,GPU and NPU tips the balance - we'll see).

What you're really not getting is that AI training can be completely separated from delivering AI services to users and there's really no downside for using (e.g.) NVIDIA hardware to develop models that are then delivered using Apple Silicon (which is very good in terms of how much CPU/GPU/NPU power it can deliver on a mobile/laptop/small-form-factr device).

And I already told you, workstations are related to servers due to how they work in terms of part.

An iPhone is "related to" a server in many ways. The sort of video and image manipulation you can run on an iPhone would have been a "workstation" job not many years ago. You can stick Apache on a MacBook Air and use it as a "server". You can turn a Mac Pro (2019 or 2023) on its side, bolt it to a rack and run PostgreSQL on it and call it a server. Heck, some years ago Apple had a product called (questionably) a "Mac Mini Server" (it was just a Mac Mini with an extra hard drive).

So, here's your problem: You can call something a "server" or "workstation" based on purely what you are using it for - but that's not what people mean when they refer to hardware that bas been designed specifically for "server" or "workstation" use. I've got a couple of Raspberry Pis that I use as "servers" - but hardware-wise they're "maker boards" using chips designed for embedded systems or set-top boxes. .

AMD make "Threadripper" processors for workstations and "Epyc" processors for servers. Yes, they're "related", yes you can run Davinci Resolve on an epyc, or SQL Server on a Threadripper - but the two products are optimised for different types of workload. Intel make half a dozen different Xeon ranges depending on whether you want a workstation (Xeon-W) or various types of server application. There's a bunch of other features that hardware specifically designed as servers often have that you wouldn't find on workstations or personal computers: lights-out power management, redundant PSUs, hot-swappable drives, rack mounting etc.

If you're buying "pro" hardware to do a specific job, you pick the hardware optimised for the task at hand - you don't buy a "workstation" to use as a "server" or vice versa. In fact, "workstation" or "server" doesn't really cut it - do you want a 3D workstation, a scientific computing workstation, an AI development workstation, a file server, a database server, a cloud compute server...

This is partly why the Mac Pro is floundering. It's a one-size-fits-all platform aimed at customers with specialist needs competing with the PC World which offers a vast range of interchangeable components (and several major operating systems) you can use to tailor a system to your exact needs. The 2019 Mac Pro priced-out anybody who just wanted a powerful desktop Mac - but only offered a very narrow range of supported CPU and GPU options c.f. the vast choice offered by PC hardware.

 

[1d1otic]

The other issue here is the increasing concerns (found all over the press) that AI companies are currently over-valued in comparison to any possible future profits, and that the AI bubble is currently fit to burst. Companies like NVIDIA and Meta are "riding the tiger" - they're already all-in on AI - and have no choice but to keep doubling down, but Apple still has the choice to maybe stay on the sidelines until the "re-adjustment" hits.

For better or worse, Apple were late to the game on AI services and AI development - they'd have their work cut out playing catchup with NVIDIA et. al. even if the current growth rate continues, but sinking investment into developing AI training hardware now runs a real risk of going up in smoke when the crash comes. NVIDIA, Meta etc. might survive, but these crashes tend to be last-in, first-out... Even if someone wanted to get a stake in the AI boom right now, I don't think it's the right time to go for the "plucky outsider" rather than someone who actually has product right now.

Apple's decision to license tech from existing AI giants, use third-party training hardware etc. rather than roll their own could prove to be very, very astute.

What Apple do have is a great platform for consuming AI services on iDevices and MacBooks with better power/performance than comparable devices. Better still, these are devices that people will still want when the petals fall off the AI tulip.


Nobody is claiming that the M2 Ultra Mac Pro is a good machine for AI training. It's not even for that - it's mainly for users who beed specialist PCIe cards that aren't GPUs/TPUs. There's already a M3 ultra and no reason not to think that there won't be a M4/M5 Ultra in due course. (There may be some cases where having 512GB of unified RAM directly accessible by CPU,GPU and NPU tips the balance - we'll see).

What you're really not getting is that AI training can be completely separated from delivering AI services to users and there's really no downside for using (e.g.) NVIDIA hardware to develop models that are then delivered using Apple Silicon (which is very good in terms of how much CPU/GPU/NPU power it can deliver on a mobile/laptop/small-form-factr device).



An iPhone is "related to" a server in many ways. The sort of video and image manipulation you can run on an iPhone would have been a "workstation" job not many years ago. You can stick Apache on a MacBook Air and use it as a "server". You can turn a Mac Pro (2019 or 2023) on its side, bolt it to a rack and run PostgreSQL on it and call it a server. Heck, some years ago Apple had a product called (questionably) a "Mac Mini Server" (it was just a Mac Mini with an extra hard drive).

So, here's your problem: You can call something a "server" or "workstation" based on purely what you are using it for - but that's not what people mean when they refer to hardware that bas been designed specifically for "server" or "workstation" use. I've got a couple of Raspberry Pis that I use as "servers" - but hardware-wise they're "maker boards" using chips designed for embedded systems or set-top boxes. .

AMD make "Threadripper" processors for workstations and "Epyc" processors for servers. Yes, they're "related", yes you can run Davinci Resolve on an epyc, or SQL Server on a Threadripper - but the two products are optimised for different types of workload. Intel make half a dozen different Xeon ranges depending on whether you want a workstation (Xeon-W) or various types of server application. There's a bunch of other features that hardware specifically designed as servers often have that you wouldn't find on workstations or personal computers: lights-out power management, redundant PSUs, hot-swappable drives, rack mounting etc.

If you're buying "pro" hardware to do a specific job, you pick the hardware optimised for the task at hand - you don't buy a "workstation" to use as a "server" or vice versa. In fact, "workstation" or "server" doesn't really cut it - do you want a 3D workstation, a scientific computing workstation, an AI development workstation, a file server, a database server, a cloud compute server...

This is partly why the Mac Pro is floundering. It's a one-size-fits-all platform aimed at customers with specialist needs competing with the PC World which offers a vast range of interchangeable components (and several major operating systems) you can use to tailor a system to your exact needs. The 2019 Mac Pro priced-out anybody who just wanted a powerful desktop Mac - but only offered a very narrow range of supported CPU and GPU options c.f. the vast choice offered by PC hardware.

Then you are justifying the limitation of Apple chips after all especially since Mac Pro WAS widely being used until Apple ditched it and now, look how poor it is claiming that M3 Ultra is the best Apple chip they can make which is awful. Again, since Apple makes their own chips, it's still important to build their own AI server with better chips cause they are using their own workflows and models just like others. You said they are also using Nvidia but you dont aware that they are starting to replace Nvidia GPU due to many limitations.

 

[tenthousandthings]

I believe that the writing is on the wall. In all likelihood, the Mac Pro will be canned very soon and Infiniband will be used to offer some limited scalability to users who need more. I doubt that Apple would bother with Infiniband if they intended to build larger or more modular systems.

On Infiniband, how do you see Apple implementing that? I think you mentioned Thunderbolt in another post (this thread is a bit out of control) — how would that work?

My understanding could be flawed, but I believe Nvidia uses Infiniband in their proprietary networking (DGX Spark has ConnectX-7 400 Gb/s, but the upcoming DGX Station will have a ConnectX-8 800 Gb/s card, and ConnectX-9 is on the roadmap at 1600 Gb/s). I’ve not heard it mentioned in connection to other non-Nvidia initiatives. What are the advantages of Infiniband over what the Ultra Ethernet Consortium is doing?

 

[leman]

On Infiniband, how do you see Apple implementing that? I think you mentioned Thunderbolt in another post (this thread is a bit out of control) — how would that work?

You can already try it out today — all you need are two Macs with Thunderbolt5 ports running the latest beta of 26.2! You connect them using a compliant cable and then you can use the industry-standard Infiniband APIs to establish the communication channel. From what I understand, the

My understanding could be flawed, but I believe Nvidia uses Infiniband in their proprietary networking (DGX Spark has ConnectX-7 400 Gb/s, but the upcoming DGX Station will have a ConnectX-8 800 Gb/s card, and ConnectX-9 is on the roadmap at 1600 Gb/s). I’ve not heard it mentioned in connection to other non-Nvidia initiatives. What are the advantages of Infiniband over what the Ultra Ethernet Consortium is doing?

This is not my area of expertise, and from doing some reading and looking at sample code it seems to me that it's important to distinguish between the logical and physical layers. Infiniband as API has been around for a while, which is presumably why Apple chose it as their software layer. At the same time, Infiniband can refer to a family of hardware networking interfaces. As I understand it, Apple does not implement the latter in their own consumer hardware. Instead, they use Thunderbolt (possibly with some proprietary sauce) to implement remote direct memory access between computers, and allow you to access this functionality in software using Infiniband protocol. So we are really talking about Infiniband over Thunderbolt, which is an Apple-specific implementation and likely has very different performance characteristics compared to the Infininaband adapters you can buy or the ones Nvidia uses.

What is not yet clear is how would connecting multiple Macs work. In principle, you'd need a switch, and I wonder whether any of the currently available TB5 switches would work with it. It is possible that two Macs is a hard limit. I'd assume that Apple will use similar technology for their proprietary servers, and the physical layer there might look very differently.

 

[theluggage]

Then you are justifying the limitation of Apple chips after all

Yes.

I've repeatedly said here and in other threads that current Apple Silicon chips are not well suited to PCIe-based workstations like the 2019 Mac Pro, and that the current Apple Silicon Mac Pro has a very limited audience who (mainly) need legacy interface cards until they're ready to move to Thunderbolt.

Apple Silicon - and the related A-series chips - are great for phones, tablets, laptops and small-form-factor systems which is where Apple make most of their money. The Mac Studio might not compete on raw horsepower with refrigerator-sized x86 workstations but it doesn't have much competition in such a small, quiet form-factor - and people are increasingly getting their "raw horsepower" on-demand from the cloud.

especially since Mac Pro WAS widely being used

"widely" in the sense of probably being Apple's lowest-selling (per unit) Mac and having negligible market share in the wider PC Workstation market.

The sort of credible PCIe workstation with discrete GPUs that you are asking for would need Apple to design a complete new PCIe-focussed CPU (which would mea throwing away many of the unique features of Apple Silicon) just for this tiny market... Without a far larger market they're simply not going to get back the huge up-front costs of designing a new processor and putting it into production. Even the current M3 Ultra has the advantage of basically being two M3 Max dies fused together - and those dies were also used in the MacBook Pro, which had a considerably larger market.

If there was a large market for the Mac Pro then Apple would have updated it more frequently, because that's the way you make money.

 

[maflynn]

If there was a large market for the Mac Pro then Apple would have updated

That's the bottom line, Apple isn't just ignoring a market just for the sake of ignoring it. If the Mac Pro was profitable, or making sufficient income, they would be investing in it

 

[dgdosen]

You can already try it out today — all you need are two Macs with Thunderbolt5 ports running the latest beta of 26.2! You connect them using a compliant cable and then you can use the industry-standard Infiniband APIs to establish the communication channel.

You'd think your garden variety Mac Youtuber would make this video instead of speculating about an M6 🙂

 

[OptimusGrime]

You'd think your garden variety Mac Youtuber would make this video instead of speculating about an M6 🙂

It depends on 26.2. I imagine once it is released, someone like Alex Ziskind will try it.

 

[leman]

You'd think your garden variety Mac Youtuber would make this video instead of speculating about an M6 🙂

I can imagine that the skill ceiling is quite high. There is no documentation, very little sample code - so you’d likely need to write and debug some obscure network code. Besides, it’s not quite clear to me how you’d make it into engaging content - it’s not like you plug two Macs together and the software jut gets faster. You’d have to code stuff for it explicitly.

There was at least one user specializing in ML content who posted some info and benchmarks on Twitter, so there’s that. I remember someone posting a link here.

 

[Basic75]

You can already try it out today — all you need are two Macs with Thunderbolt5 ports running the latest beta of 26.2! You connect them using a compliant cable and then you can use the industry-standard Infiniband APIs to establish the communication channel.

Wait, there's an implementation of the infiniband verbs api using Thunderbolt as transport? Does it support rdma or emulate it? And where can I find this?

 

[OptimusGrime]

Wait, there's an implementation of the infiniband verbs api using Thunderbolt as transport? Does it support rdma or emulate it? And where can I find this?

You could start here:

https://forums.macrumors.com/threads/apple-silicon-deep-learning-performance.2319673/page-17?post=34304159#post-34304159

 

[leman]

Wait, there's an implementation of the infiniband verbs api using Thunderbolt as transport? Does it support rdma or emulate it? And where can I find this?

The measured latency is allegedly in microsecond scale, so it’s very likely real RDMA. You can check out the infiniband/verbs.h header shipped with latest macOS and Xcode betas.

 

[mcnallym]

M2 Ultra is too slow to use cause Apple has no choice but to use their own chips other than Nvidia. They are not even close to TPU servers. And I already told you, workstations are related to servers due to how they work in terms of part. Why do you keep justifying poor hardware for? If you think Mac Studio or Ultra chips are enough, then you are short sighted and will only kill Apple's future.

Clearly, Apple is too limited with hardware resources.

why does Apple have no choice but to use there own chips?

Clearly Apple has the choice to use non Apple Products.

I have previously given you a list of products that Apple used to make but no longer does yet do they no longer have the requirements as a compnay for the products that they don’t make.

So if Apple has to use its own products as you say then how does Apple

do networking
do storage
do printing
do backups
do servers

It would appear that Apple does indeed have the choice to use non-Apple products where doesn’t make a suitable product and actively does use non Apple products.

So why is AI hardware any different other then YOU WANT it to be.

I am not justifying poor hardware, I am telling you that you don’t need to develop a workstation, ie a Mac Pro to develop a Chip used for AI Training.

Google has not developed a Workstation
Nvidia has not developed a Workstation

So why when these companies not developed a workstation but have there own AI hardware for attaining is that Apple needs to develop a workstation.

I am telling you that Apple don’t have to develop their own hardware to do AI Training and that if they did develop there own hardware you wouldn’t start with a Mac Pro Chip.

Do you actually know what the M2 Ultra machines are being used for? They are not used for AI Training. They are being used to provide Apple Intelliegence SERVICE to End Users that not suited to being completed on the End User device. For which they do the job being asked to do.

All you have at the moment is YOU want a Mac Pro in the mold of the 2006-2012 models, and then trying to make a case for it that Apple need one so that can do AI Training. Yet have failed to bring anything to actually make the case.

 

[maflynn]

why does Apple have no choice but to use there own chips?

Clearly Apple has the choice to use non Apple Products.

To use non-apple products - yes, makes a lot of sense. To sell computers using non-apple components (where they offer apple components), that's business suicide. They've heavily marketed their apple silicon as superior to anything out there, using a competitor's component would be a horrible look for apple.

If they were to sell a computer that's using AMD (CPU or GPU) or Nividia's (GPU), they're all but admitting apple silicon is inferior, which would be disastrous

 

[mcnallym]

To use non-apple products - yes, makes a lot of sense. To sell computers using non-apple components (where they offer apple components), that's business suicide. They've heavily marketed their apple silicon as superior to anything out there, using a competitor's component would be a horrible look for apple.

If they were to sell a computer that's using AMD (CPU or GPU) or Nividia's (GPU), they're all but admitting apple silicon is inferior, which would be disastrous

Bit confused as the OP is saying Apple have to develop their own hardware, as in Apple does not have their own product for the job of AI Training.
Not aware where anyone said they should a non apple product where they have an apple product.

Yes using an Nvidia or AMD GPU in an End User Workstation undermines Apple Silicon but I certainly didn’t say that should be done.
Apple Silicon already does what it needs to for what Apple DESIGN it to be used for as well as that would break the UMA that use as part of the system.

Apple trains Apple intelligence on Google’s Tensor TPU alongside using own ASi systems to fill in the parts that Apple Silicon doesn’t do. Main thing is though that is NOT something Apple is selling to a user and clearly the Tensor TPU does the work the Apple Silicon doesn’t.

However that is not an end user product and they don’t have Google tensor TPU in the user devices running Apple Intelligence as Apple Silicon has everything needed in the end device.

Perhaps I should have posted, why does Apple have to develop their own product for every requirement that they have in-house instead however I figured that to use their own product they would have to have developed the product in the first place so was easier just to type as I did.

 

[tenthousandthings]

So if Apple has to use its own products as you say then how does Apple

do networking
do storage
do printing
do backups
do servers

It would appear that Apple does indeed have the choice to use non-Apple products where doesn’t make a suitable product and actively does use non Apple products.

Apple also doesn’t do display panels, a critical component of their business that appears in most of their high-volume products. If we’re looking for the next area where Apple could easily become frustrated with the status quo, it’s probably that (maybe a joint venture, but called “Apple Display”), and not $100,000 AI workstations that run Ubuntu/Linux.

 

[1d1otic]

why does Apple have no choice but to use there own chips?

Clearly Apple has the choice to use non Apple Products.

I have previously given you a list of products that Apple used to make but no longer does yet do they no longer have the requirements as a compnay for the products that they don’t make.

So if Apple has to use its own products as you say then how does Apple

do networking
do storage
do printing
do backups
do servers

It would appear that Apple does indeed have the choice to use non-Apple products where doesn’t make a suitable product and actively does use non Apple products.

So why is AI hardware any different other then YOU WANT it to be.

I am not justifying poor hardware, I am telling you that you don’t need to develop a workstation, ie a Mac Pro to develop a Chip used for AI Training.

Google has not developed a Workstation
Nvidia has not developed a Workstation

So why when these companies not developed a workstation but have there own AI hardware for attaining is that Apple needs to develop a workstation.

I am telling you that Apple don’t have to develop their own hardware to do AI Training and that if they did develop there own hardware you wouldn’t start with a Mac Pro Chip.

Do you actually know what the M2 Ultra machines are being used for? They are not used for AI Training. They are being used to provide Apple Intelliegence SERVICE to End Users that not suited to being completed on the End User device. For which they do the job being asked to do.

All you have at the moment is YOU want a Mac Pro in the mold of the 2006-2012 models, and then trying to make a case for it that Apple need one so that can do AI Training. Yet have failed to bring anything to actually make the case.

Nvidia DOES have workstations and it calls DGX.

And again, Apple is only limiting their own hardware so that many professional apps are either gone or restricted. Do you really think that's normal? Since Apple needs AI more than ever instead of idiotic Apple Intelligence, it's just a matter of time before they make their own severs with their own chips.

You might say why then tell that to many companies using their own chips for their own models and workflows such as Google, Meta, Amazon, Microsoft, Tesla, and more.

 

[leman]

Nvidia DOES have workstations and it calls DGX.

Are you referring to the new Spark? That’s literally a clone of Mac Studio, only with an anemic CPU, slow RAM, and a rebranded mobile 5070…

I don’t really follow you here. Are you telling us that Apple should make computers like Spark? They’ve been doing it years before Nvidia.