Poll: When will there be an arm mac that exceeds a 2019 mac pro in performance?

[konqerror]

Intel has been too busy resting on it's laurels to care about any particular market segment, "you want workstation chips or single socket servers? then take this Xeon & make do, we ain't got no time for that"...

There's real production costs. Intel is saying that the costs of a mask set and the engineering costs to delete a bunch of unused cores is more than the cost of throwing away half or more of the CPU die. Think about that.

I stand by my statement 100%. Either Apple will sell somebody else's CPUs for the Mac Pro or the product will be killed. The costs simply make no sense for a product very few people buy.

I don't know why people have a knee-jerk fanboy reaction about this. Apple simply buys the chips from Fujitsu and says that we're using the same CPUs as the world's fastest supercomputer. Marketing done.

 

[Realityck]

• To me the bigger question is can they satisfy what we have today in a meaningful way.
• They have far more control now. What does that look like long term? I don’t know, I don’t think anyone of us do.

This is two edged sword, true they can maximize design goals offering a paired OS to hardware, but at the same time, if Apple lacks competition that keeps pushing them forward you just end up what they want to sell as a computer.

One has to wonder if you will get what you really want in the end because perhaps Apple wants to take this decision away from you, referring to your early comment about being satisfied in a meaniful way. Apple has done this in the past, computer in a box with everything they think you need. Look at their design history.

I like it when Apple is having motivation for its users, not motivate for its shareholders. Their past keynotes are famous for everything sounds awesome, but is it?

 

[burgerrecords]

The *entire point* of Apple investing in their own silicon designs is that they control the entire stack top to bottom. They’re simply not going to outsource a higher end design.

this is what Apple wants you to believe, that only their magic synergies happen with vertical integration - when clearly there’s a lot fewer meaningful difference between Apple products and commodity products from the beginning of time to today. From the velocity engine to the neural engine it’s largely branding.

there are a few meaningful differences though and the combination of tight control and massive financial wherewithal and enthusiastic Mac users lends itself to be able to pull this move to what is a nice move toward better portable computer that slot into their much more important iOS user base.

All things equal, open standard computing appeals to me personally more.

 

[Waragainstsleep]

As I said before, Apple does not sell the individual chips. And they can certainly afford it to subsidize a niche high-visibility product with strong psychological impact like Mac Pro

If an Apple Silicon Mac Pro offers something different and desirable compared to the Intel powered offerings from Dell and HP, it will be much less of a niche product. There is no shortage of takers when it comes to buying the best equipment possible.

I stand by my statement 100%. Either Apple will sell somebody else's CPUs for the Mac Pro or the product will be killed. The costs simply make no sense for a product very few people buy.

This might be plan B at a push but the money if Apple spend it will be classed as R&D and the tricks they learn in pulling off such a product will pay dividends down the line as those features filter down to consumer products through optimisation and efficiency improvements as has always happened with cutting edge tech. Apple is happy to play the long game and they can afford to do so way more than Intel can. Intel shanking their engineering budget on a venture that might not work and might set their core business back years is a gamble they won't take, especially given they are already years behind the curve. At least they are perceived to be. And perceptions matter to companies with conventional boards and shareholders like Intel.
Apple don't have that concern because their management has the faith and credibility from their track record to do whatever the **** they think is a good idea. And for every idea that doesn't pay off, there are more than enough that do. Again I refer you to the Apple Car.
And if this one doesn't pay off, they can always buy a few chips from another vendor to tide them over.

I don't know why people have a knee-jerk fanboy reaction about this. Apple simply buys the chips from Fujitsu and says that we're using the same CPUs as the world's fastest supercomputer. Marketing done.

Will those Fujitsu chips support all the features in Mac OS that the consumer models do? Plenty of engineering work to integrate the T2 with a new CPU from scratch.

If Apple doesn't manage to make a 128 Core CPU within two years, they can always take whatever kick-ass chip they will already be using in the iMac and drop two of them on one Mac Pro board. Or four. They have options and you can bet they will have explored them thoroughly. I wouldn't be surprised if they already have something that's all but ready to go.

this is what Apple wants you to believe, that only their magic synergies happen with vertical integration - when clearly there’s a lot fewer meaningful difference between Apple products and commodity products from the beginning of time to today. From the velocity engine to the neural engine it’s largely branding.

The certainly was the case thus far but they have made the integration of hardware and software work very well for them ever since OS X. That vertical integration has worked wonders for stability, security and design and they are just going the extra mile now to make sure they can design the things they really want to.
The last iMac G5 was supposedly a rush job because Intel weren't ready in time, then the 32-bit Core Duo was swiftly replaced so its no stretch that they were originally planning to kick off Intel Macs with 64-bit across the board. Like their 64-bit G5s. They ended up having to compromise and cut corners because of Intel. That doesn't happen any more. They make their own chips, and they make their own contingencies and this lets them make their own timetables.

there are a few meaningful differences though and the combination of tight control and massive financial wherewithal and enthusiastic Mac users lends itself to be able to pull this move to what is a nice move toward better portable computer that slot into their much more important iOS user base.

All things equal, open standard computing appeals to me personally more.

To be fair they've long been quite good at supporting open standards. Even beyond what they admitted to.

 

[konqerror]

Will those Fujitsu chips support all the features in Mac OS that the consumer models do? Plenty of engineering work to integrate the T2 with a new CPU from scratch.

The T1 and T2 fundamentally connect to the system via USB, exactly like an iPhone that's been plugged in. It also emulates a standard flash device via SPI. Nothing would require any custom engineering work.

If Apple doesn't manage to make a 128 Core CPU within two years, they can always take whatever kick-ass chip they will already be using in the iMac and drop two of them on one Mac Pro board. Or four. They have options and you can bet they will have explored them thoroughly. I wouldn't be surprised if they already have something that's all but ready to go.

Here you're significantly underestimating the problem. You have to design a high speed cache coherent external interface. Then you have to implement directories and snooping for cache performance. The ring bus can't handle this traffic, so now it's a mesh.

This is why the Xeons are very different from desktop systems. The entire external interface of the core has been redesigned.

If you want to see how hard things are, people should look at the complexity in existing systems instead of making wild guesses.

 

[ian87w]

This is two edged sword, true they can maximize design goals offering a paired OS to hardware, but at the same time, if Apple lacks competition that keeps pushing them forward you just end up what they want to sell as a computer.

One has to wonder if you will get what you really want in the end because perhaps Apple wants to take this decision away from you, referring to your early comment about being satisfied in a meaniful way. Apple has done this in the past, computer in a box with everything they think you need. Look at their design history.

I like it when Apple is having motivation for its users, not motivate for its shareholders. Their past keynotes are famous for everything sounds awesome, but is it?

I disagree. The whole PC industry remains as a valid competitor. In fact, it's pushed Apple this far into investing into their own silicon. And don't forget the elephant in the room, Qualcomm. Companies like Microsoft, Samsung, etc are probably itching to also join the likes of Apple into a non-x86 market.

 

[burgerrecords]

The certainly was the case thus far but they have made the integration of hardware and software work very well for them ever since OS X.

I would argue the last time integration *really* mattered was up *until* OS X. I owned no System X.X macs and during my 1983 through 1998 DOS and windows days and it could get really harry on those platforms (there was real contrast between "just working" on Mac and most people's experience - like mine). By the time the apple were selling a fully Windows compatible computer themselves, Windows was very good at allowing infinite hardware and software configurations (especially decent quality hardware and software) to run well.

In 2020 Linux is extremely stable and secure (more so than Mac) and most hardware "just works" too. Windows 10 is fine if you have a corporate managed system (where most of microsofts 1.5 trillion value comes from) or know what you are doing. I agree Mac is still best for everyone else who isn't on a tight budget (it's not like we're going to see Macs used in developing countries by anyone but the elites)

To be fair they've long been quite good at supporting open standards

Not in the way someone who prefers openness would wish. I would prefer to user my Airpods on my windows 10 computer, or to send and receive iMessages from a non Apple device. Again i think closed systems were justified when personal computing was in its infancy, but today with Apple it is largely about lock in (which is fine i think people know what they're getting into) And perhaps from a pure consumer standpoint Apple is close to perfect as is. For enthusiasts or the frugal it's not all roses.

I would love to see the obviously talented UI folks at Google (i'm biased as a friend of mine on those teams) create a fully realized linux distribution for a full powered Google PC. So much of the hard work is done, and it could that balance of stability, security, openness and polish that is still missing the marketplace. Obviously i'm selling my soul to google to get something like this so there's that trade off - but it would push the industry and there would like be ways to de-google a linux/google OS.

 

[Waragainstsleep]

I would argue the last time integration *really* mattered was up *until* OS X. I owned no System X.X macs and during my 1983 through 1998 DOS and windows days and it could get really harry on those platforms (there was real contrast between "just working" on Mac and most people's experience - like mine). By the time the apple were selling a fully Windows compatible computer themselves, Windows was very good at allowing infinite hardware and software configurations (especially decent quality hardware and software) to run well.

It certainly used to be the case that Macs ran Windows better than other Windows PCs. The reason typically being because Apple wrote drivers for every component and tested them together where other makers just got everything off the shelf and didn't fully test the whole experience. Thats where the integration comes in.

In 2020 Linux is extremely stable and secure (more so than Mac) and most hardware "just works" too. Windows 10 is fine if you have a corporate managed system (where most of microsofts 1.5 trillion value comes from) or know what you are doing. I agree Mac is still best for everyone else who isn't on a tight budget (it's not like we're going to see Macs used in developing countries by anyone but the elites)

Linux is stable and secure but nowhere near as usable and neither are as pretty. I think most of Microsoft's business these days is services isn't it?

Not in the way someone who prefers openness would wish. I would prefer to user my Airpods on my windows 10 computer, or to send and receive iMessages from a non Apple device. Again i think closed systems were justified when personal computing was in its infancy, but today with Apple it is largely about lock in (which is fine i think people know what they're getting into) And perhaps from a pure consumer standpoint Apple is close to perfect as is. For enthusiasts or the frugal it's not all roses.

Their model of not charging for software means that sometimes they have to keep certain features to themselves. iMessages would not be assure if you could run it on Windows or Android. Again, integration is a big factor there because its likely that the stability of the app would be significantly less on systems where Apple had not tested it with other hardware and software and crashes tend to be the starting point of vulnerabilities. It wouldn't be as good a product and why would Apple devote the resources to secure it just to add features to rival platforms? Doesn't make a lot of sense where security is a big part of the product.

I would love to see the obviously talented UI folks at Google (i'm biased as a friend of mine on those teams) create a fully realized linux distribution for a full powered Google PC. So much of the hard work is done, and it could that balance of stability, security, openness and polish that is still missing the marketplace. Obviously i'm selling my soul to google to get something like this so there's that trade off - but it would push the industry and there would like be ways to de-google a linux/google OS.

The only reason Google would this is to capture all your data. Defeats the object for a lot of Linux users doesn't it? Standards are great but companies exist to make money, its a wonder there are as many open standards as there are.

 

[Waragainstsleep]

The T1 and T2 fundamentally connect to the system via USB, exactly like an iPhone that's been plugged in. It also emulates a standard flash device via SPI. Nothing would require any custom engineering work.

The T2 incorporates the SSD controller among other things. That doesn't sound like something you'd just plug in via USB. Are you sure you know what you're talking about here?

Here you're significantly underestimating the problem. You have to design a high speed cache coherent external interface. Then you have to implement directories and snooping for cache performance. The ring bus can't handle this traffic, so now it's a mesh.

This is why the Xeons are very different from desktop systems. The entire external interface of the core has been redesigned.

If you want to see how hard things are, people should look at the complexity in existing systems instead of making wild guesses.

The point is, dual CPU boards have existed a long time and Apple has had a decade or more to work on this hardware. Claiming they can't do something because "its hard" is just as wild a guess at this point. They've overtaken the industry leaders in that time for performance and power consumption. According to the documentation, your USB T2 is actually a coprocessor so there's clearly some meshing going on there.

 

[burgerrecords]

The only reason Google would this is to capture all your data. Defeats the object for a lot of Linux users doesn't it?

Data capture and openness are different categories. Yes linux users prefer both, but openness overall is more important right to Linus himself. Do you use google? I dislike it in principal just like i dislike much of apple's approach in principal but they are both good; both are evil in different ways. Google actually wants your data but doesn't want to tie to you personally, ideally - it's an interesting question what the implications of what they know are.

But my primary point is that it's a bit tired that Apple needs to be as closed for things to work correctly or have magically synergies between hardware and software. If it's primarily to make money then let's just say that. I prefer openness.
[automerge]1595827477[/automerge]

integration is a big factor there because its likely that the stability of the app would be significantly less on systems where Apple had not tested it with other hardware and software and crashes tend to be the starting point of vulnerabilities

i just don't really see this is how it works anymore as someone who uses linux a bit, mac some, lot's of iOS and Apple TV, windows a lot.

 

[leman]

I don't know why people have a knee-jerk fanboy reaction about this. Apple simply buys the chips from Fujitsu and says that we're using the same CPUs as the world's fastest supercomputer. Marketing done.

You are not looking at the available facts

- Apple CPUs most likely have higher IPC than A64X or other HPC chips

- HPC or server chips have ultimately a completely different performance profile than a workstation. They are optimized for parallel throughout while a workstation needs to support hybrid asymmetric workloads

- Apple CPUs implement a later ARM ISA different standard than A64X. Apple CPUs contain custom ISA extensions (AMX). Finally, I am sure that Apple CPUs contain specialized instructions to support Rosetta 2. You can’t just modify an A64X to include all that. And Apple wouldn’t choose to fragment their own hardware where they have the chance - for the first time ever - to
unify its capabilities

- finally, again: the Mac Pro does not have to be profitable. It’s a show off product. The PR and indirect revenue will more than offset the cost of chip design

 

[konqerror]

You are not looking at the available facts

You have a number of serious misconceptions

- Apple CPUs most likely have higher IPC than A64X or other HPC chips

A conjecture. Again, scaling to large core counts necessarily decreases effective IPC due to bandwidth constraints and memory latency going way up. Intel has the same problem.

Again, you have to do a lot of work to make sure the IO system can feed the same cores. This is why it's not an issue of pasting 64 iPhone cores together. A ton of engineering needs to happen.

- HPC or server chips have ultimately a completely different performance profile than a workstation. They are optimized for parallel throughout while a workstation needs to support hybrid asymmetric workloads

Wrong. You can see this on Intel. The instruction latency and pipelines on the world's fastest Xeon supercomputer are identical to that of the Macbook Air. Same core.

You are imagining somehow the architecture changes. It does not. Everything around the core changes, which is what I've been repeatedly saying.

You don't understand large CPUs. It's never a core issue, its an I/O issue.

- Apple CPUs implement a later ARM ISA different standard than A64X. Apple CPUs contain custom ISA extensions (AMX). Finally, I am sure that Apple CPUs contain specialized instructions to support Rosetta 2. You can’t just modify an A64X to include all that. And Apple wouldn’t choose to fragment their own hardware where they have the chance - for the first time ever - to
unify its capabilities

Wrong. ARM is notoriously strict about binary compatibility. You cannot touch the ISA. As I said, Apple's accelerators are a CEVA block outside the CPU. That's like saying an Nvidia GPU on a computer is somehow affecting the Intel ISA.

And they've already "fragmented" their ISA. Sort of how iPhone 11 programs don't run on the iPhone 7, you know since there's huge binary incompatibilities. Oh wait...

- finally, again: the Mac Pro does not have to be profitable. It’s a show off product. The PR and indirect revenue will more than offset the cost of chip design

I'm saying the Mac Pro needs to not cost 3x it's price, which is what you're suggesting by having Apple engineer its own large core count CPU. People already complain about how much the wheels cost.
[automerge]1595833787[/automerge]

The point is, dual CPU boards have existed a long time and Apple has had a decade or more to work on this hardware. Claiming they can't do something because "its hard" is just as wild a guess at this point. They've overtaken the industry leaders in that time for performance and power consumption.

I never said they can't. I said it's too expensive and a distraction for a niche product. Especially when they can buy.

Apple has not made a dual socket Mac since 2012.

You're saying it's cheap and easy. The whole industry says it's not.

According to the documentation, your USB T2 is actually a coprocessor so there's clearly some meshing going on there.

Apple marketing for fanboys. Apple documented how it acts. It's a USB device and SPI.

 

[leman]

Wrong. ARM is notoriously strict about binary compatibility. You cannot touch the ISA. As I said, Apple's accelerators are a CEVA block outside the CPU. That's like saying an Nvidia GPU on a computer is somehow affecting the Intel ISA.

Your information is outdated. ARM officially allowed it's users to implement custom ISA extensions last year. Apple uses it's own NPU IP since A12. Apple A12 CPU cores have also introduced specialized DSP hardware and SIMD instructions called AMX

Also, Apple implements ARMv8.4-A on their CPUs, Fujistu A64FX for example is ARMv8.2-A

Again, you have to do a lot of work to make sure the IO system can feed the same cores. This is why it's not an issue of pasting 64 iPhone cores together. A ton of engineering needs to happen.

Precisely. And given Apple's expertise in cache and interconnect designs, plus the fact that they sit on a large pile of cash, I see no reason why they can't pull it off.

Wrong. You can see this on Intel. The instruction latency and pipelines on the world's fastest Xeon supercomputer are identical to that of the Macbook Air. Same core.

Exactly. Same core. Different performance profile configuration. Intel server CPUs are configured to trade peak single-core performance for sustained parallel performance. The W-class CPUs Apple uses offer both high single-core peaks and good sustained performance. The ARM server CPUs you like to mention are optimized for parallel throughput. Different use case, different profile.

I'm saying the Mac Pro needs to not cost 3x it's price, which is what you're suggesting by having Apple engineer its own large core count CPU. People already complain about how much the wheels cost.

You are tunnel visioning this. I will repeat again: the Mac Pro itself does not have to be profitable. Apple can afford to subsidize the development of workstation-class chip from it's other profits. You are thinking in terms of pure momentary profits and ignoring the psychological or ideological factors. Apple will not announce a move to it's int-house designed chips and then buy someone else's hardware, especially one that does not support Apple's latest features and does not integrate well with their design vision.

 

[konqerror]

Your information is outdated. ARM officially allowed it's users to implement custom ISA extensions last year.

Incorrect. ARM has had a coprocessor interface since the Acorn days. That's not fundamental to the ISA. You call it with a separate interface, broadly analogous to IO ports on x86, just like if you had different video cards or hardware back then.

Precisely. And given Apple's expertise in cache and interconnect designs, plus the fact that they sit on a large pile of cash, I see no reason why they can't pull it off.

Apple has released zero interconnects. The iPhone is licensed ARM AMBA and MIPI IP. Again, you don't know what you're talking about.

Exactly. Same core. Different performance profile configuration.

You're just spewing pseudo-technical BS at this point. I'm done.

 

[leman]

Incorrect. ARM has had a coprocessor interface since the Acorn days. That's not fundamental to the ISA.

Well, look it up. They announced it at last years ARM TechCon. You are so sure of your own infallibility that you don't even look up the information.

You're just spewing pseudo-technical BS at this point. I'm done.

Turbo boost profile is pseudo-technical BS? I wish

Go compare the clock setups of Xeon SP and Xeon-W and then we can talk.

 

[konqerror]

Well, look it up. They announced it at last years ARM TechCon. You are so sure of your own infallibility that you don't even look up the information.

You're getting marketing BS mixed up with technical knowledge. I know what they're talking about. You evidently didn't know anything about ARM coprocessors.

Go compare the clock setups of Xeon SP and Xeon-W and then we can talk.

As more evidence that you don't know what you're talking about, a Xeon W is a Xeon SP XCC die. No difference.

The SPs have different models which emphasize one over the other. The hardware is absolutely the same, it's a binning issue.

Also, Apple implements ARMv8.4-A on their CPUs, Fujistu A64FX for example is ARMv8.2-A

No they don't A12 is 8.3, A11 is 8.2. It doesn't affect application software.

In fact, Apple rejects any software not compiled for ARMv8.0-A

Preparing your app to work with pointer authentication | Apple Developer Documentation

Test your app against the arm64e architecture to ensure that it works seamlessly with enhanced security features.

developer.apple.com

Just like saying because Apple is shipping Coffee Lake CPUs, then all your Skylake and Ivy Bridge Macs are completely useless. Because there's real instruction changes on those.

Another example of marketing.

 

[Waragainstsleep]

Data capture and openness are different categories. Yes linux users prefer both, but openness overall is more important right to Linus himself. Do you use google? I dislike it in principal just like i dislike much of apple's approach in principal but they are both good; both are evil in different ways. Google actually wants your data but doesn't want to tie to you personally, ideally - it's an interesting question what the implications of what they know are.

I'm quite sure Google wants to tie your data to you. Its outside forces that prevent them from doing as they'd like. Theres a certain amount of (American) paranoia when it comes to a love of openness. This for me does not mesh with Google's level of note-taking.

But my primary point is that it's a bit tired that Apple needs to be as closed for things to work correctly or have magically synergies between hardware and software. If it's primarily to make money then let's just say that. I prefer openness.
[automerge]1595827477[/automerge]

i just don't really see this is how it works anymore as someone who uses linux a bit, mac some, lot's of iOS and Apple TV, windows a lot.

Just because you're over it doesn't stop it being true. Theres definitely an element of truth to it at the least. Apple doesn't scream about the fact that its there to make money because who does? But they have never tried to deny it either. The vast majority of Google users have no clue about what Google keeps or how their business model works. They see them as a generous benevolent entity that gives everyone cool stuff.
iMessage is probably not the killer feature it was a few years ago (or maybe should have been), but its key selling point was its encryption and that would defeat the object entirely for Google hence there was never an equivalent for Android.

I never said they can't. I said it's too expensive and a distraction for a niche product. Especially when they can buy.

Theres engineering to be done even if they buy. It seems unlikely though I expect they have prototypes running. Just in case.

Apple has not made a dual socket Mac since 2012.

But they have made them. With two completely different CPU families.

You're saying it's cheap and easy. The whole industry says it's not.

I'm saying its possible and therefore its an option. Apple has never been scared of expensive and difficult.

Apple marketing for fanboys. Apple documented how it acts. It's a USB device and SPI.

The Apple documentation says its the SSD controller, encrypts data on the SSD, accelerates video and audio encoding, handles Hey Siri, Touch ID and more. That sounds like a lot for a UBS device. By all means show us the documentation though.

 

[leman]

You're getting marketing BS mixed up with technical knowledge. I know what they're talking about. You evidently didn't know anything about ARM coprocessors.

On the TechCon they were talking about reserving ARM ISA space for "partners". I don't care wether there is a coprocessor involved or not. We are talking about custom ARM instructions, in words of ARM representatives themselves. And it is a fact that Apple A12 has custom instructions, I linked you to a thread on realworldtech. So your standpoint that ARM does not allow their licensees to create their own instructions is very odd: we have direct evidence to the contrary.

As more evidence that you don't know what you're talking about, a Xeon W is a Xeon SP XCC die. No difference.

The SPs have different models which emphasize one over the other. The hardware is absolutely the same, it's a binning issue.

You are not listening. Yes, they are the same hardware. But they are configured for different niche. A Xeon W has higher single-core turbo to better support hybrid workloads. Scalable Xeons are more intended as a pure multiprocessing solution.

And this is exactly the point. You mention the HPC ARM CPUs that Apple could use. Let's take the Fujitsu A64FX. It's a 48 core beast with focus in numerical computing (with 512bit SVE). This chip is built for parallel workflows. It's a supercomputer chip. It doesn't seem to offer dynamic overclocking (aka. turbo boost) and it is ridiculously expensive ($40000 for a dual-CPU box from I could find). How is this a suitable chip for Apple's needs? Or do you really think that licensing Fujitsu core design, making a smaller chip on its base, adding all the Apple extensions, integrating it with Apple's power management and making sure that it offers adaptive performance is somehow easier than taking Apple's already excellent core design and scaling it up to a 20+ cores? Or that it's going to be cheaper?

No they don't A12 is 8.3, A11 is 8.2. It doesn't affect application software.

In fact, Apple rejects any software not compiled for ARMv8.0-A

A13 is 8.4

And yes, Apple only accepts ARMv8.0 code right now so that all users can reasonably run all the apps - for iOS targets. Who says that Mac targets are going to be the same? That's new hardware with new unified set of capabilities.

Apple has released zero interconnects. The iPhone is licensed ARM AMBA and MIPI IP. Again, you don't know what you're talking about.

No, I'm not. I am not a hardware engineer. I am a software developer with interest in CPU architecture and micro-optimisations. I don't have any insider knowledge about Apple interconnects. All I know on the matter is from reading technical analysis from people more knowledgeable than me. In-depth analysis of A13 cache system suggests that it is state of the art and competitive with current desktop offerings.

If you say that Apple is licensing interconnect tech for their SoC — I believe you. You clearly know what you are talking about. But I don't see why it matters. Apple will either develop interconnect suitable for larger chips or they will license it from someone. All of these are solvable challenges, and Apple has the political will, the money and the human resources to pull it off (by the they have been actively hiring interconnect engineers lately). Definitely more realistic than licensing an extremely expensive HPC supercomputing chip and somehow hammering it into a multipurpose workstation...

 

[Waragainstsleep]

https://www.fujitsu.com/global/Images/tofu-interconnect2.pdf

If Fujitsu can make their own interconnect tech, what makes you think Apple can't?

 

[leman]

A conjecture. Again, scaling to large core counts necessarily decreases effective IPC due to bandwidth constraints and memory latency going way up. Intel has the same problem.

P.S. About this — here is an official slide that mentions "near-Xeon integer performance". Well, Appel already exceeded Xeon in integer performance. Which again reinforces what I have been repeating again and again — the purpose of the Fujitsu monster is HPC. It's "gimmick" is a lot of parallel cores with 512-bit FP vector ALUs. This is not a general purpose CPU and it's a poor fit for a workstation.

 

[JMacHack]

The costs simply make no sense for a product very few people buy.

1. Apparently you fail to understand the point of the Mac Pro, if this was the case, then they wouldn't have made one at all.
2. Margins, I'd bet the margins on the Mac Pro are higher than the margins on any other Apple product. All workstation-class stuff has high margins.
3. Following that point, so do Xeon-class chips. Apple would likely save money by printing out workstation-class chips vs. buying them from Intel (or any other manufacturer) which would increase their margins.
4. Why would Apple need to buy (or even pay another company to design chips for them, lol) when they could literally do it themselves? We've seen that their in-house designs are world-class. even to the point that their passively-cooled iPhone and iPad chips are nipping at the heels of lower-mid range laptops and low-range desktops. Especially when they're designing chips for every other Mac.

Your cost-benefit analysis doesn't even make a bit of sense.

Even if you assume that the Mac Pro-level chip must be a clean-sheet design (which it doesn't, it can easily share architectual elements with the other Mac Silicon) ordering custom-made chips from another party makes even less sense.

And calling us "fanboys" doesn't make your argument any more valid.

 

[konqerror]

It's "gimmick" is a lot of parallel cores with 512-bit FP vector ALUs.

You just described a Xeon W. What's AVX-512?
[automerge]1595872509[/automerge]

If Fujitsu can make their own interconnect tech, what makes you think Apple can't?

That's the result of billions of dollars of Japanese government funded supercomputer sales. That was the same interconnect carried over from the SPARC line that made world's fastest supercomputers multiple times.
[automerge]1595872597[/automerge]

And calling us "fanboys" doesn't make your argument any more valid.

Pretty much nobody on this thread has said anything sensible about computer architecture. Somebody just denied the existence of AVX-512 on Intel processors.

 

[leman]

That's the result of billions of dollars of Japanese government funded supercomputer sales. That was the same interconnect carried over from the SPARC line that made world's fastest supercomputers multiple times.

Billions of dollars? I think I can see Tim Cook getting his wallet out...

 

[JMacHack]

Pretty much nobody on this thread has said anything sensible about computer architecture. Somebody just denied the existence of AVX-512 on Intel processors.

Your entire argument hinges on Apple not being able to create an interconnect.

 

[konqerror]

Your entire argument hinges on Apple not being able to create an interconnect.

Your entire belief that it will happen is that Apple wants to completely rearchitect cache, on-chip, and potentially external interconnects for a niche product that sells less units a year than iPhones in 1-2 days.

Particularly when multiple vendors sell the exact product you need.

Everyone arguing completely misestimates the cost of CPU engineering and how niche the Mac Pro is.