No “Extreme” chip coming to Mac Pro?

[theorist9]

Come on dude, there are enough Mac Pro wheel jokes over at the Other Place, no need to descend to that level...

It wasn't a simple wheel joke, it was a multi-layered joke about stuffing unnecessary features into the ASD rather than focusing on the fundamentals...while, yes, also taking another swipe at the wheels. Indeed, that's why I specifically chose the A13, and also deliberately borrowed their marketing language, substituting "innovative wheel features" for "innovative Studio Display features":

I figured this was the type of crowd that would appreciate more layered humor.

 

[leman]

Perhaps one advantage of an isolated dGPU is freedom of sloppiness. The CPU sends the code (pretty compact), description files (still pretty small) and textures (generally less small) and the GPU chomps on it to produce the result. If the output is to the screen, the GPU just handles it and the processor goes on doing what it was doing, unaffected (assuming the card includes the display driver).

Ultimately, the data transfer is usually a lot smaller than the work frame, and if the GPU has enough room, it can generate intermediate content with abandon (such as hidden surfaces that do not sow up in the final output) – sloppiness. The final result (image) will still be smaller to pass back, if necessary, than the GPU's overall workset. I believe the same is somewhat true for heavy math jobs, though somewhat less so.

So a UMA iGPU has to be a little more elegant in how it runs, making more judicious use of limited memory space and bus bandwidth. The code has to be more efficient, which means that an iGPU will almost always draw less power for the same job, because it will do it differently. It would seem that the dGPU is the sledgehammer approach.

If the iGPU has significantly lower bandwidth than the dGPU, yes, bandwidth optimisations are key. There is a good reason why mobile GPUs are almost exclusively tilers that use all kinds of tricks to avoid fetching stuff from RAM. But this is less of a factor for Apple Silicon, which has reasonably high memory bandwidth and large caches. Of course, all GPUs massively benefit from bandwidth and caching optimisations (there is a reason why Nvidia has been heavily investing in compute data compression techniques as well as tiling), and Apple does have some unique advantages in this area. The big selling point of iGPUs is that they make system architecture simpler. There is only one pool of RAM to worry about. No need to shuffle memory back and forth and track what's where.

However, Apple still has some work to do in the GPU department. It would be great if they could offer unified virtual memory in addition to the unified physical memory. There is still some substantial degree of manual work that needs to be done to track GPU resources.

 

[B01L]

It wasn't a simple wheel joke, it was a multi-layered joke about stuffing unnecessary features into the ASD rather than focusing on the fundamentals...while, yes, also taking another swipe at the wheels. Indeed, that's why I specifically chose the A13, and also deliberately borrowed their marketing language, substituting "innovative wheel features" for "innovative Studio Display features":

View attachment 21560

I figured this was the type of crowd that would appreciate more layered humor.

Sorry, just waded thru some of the newer ASi Mac Pro threads and was exposed to too many wheel jokes...

 

[Yoused]

However, Apple still has some work to do in the GPU department. It would be great if they could offer unified virtual memory in addition to the unified physical memory. There is still some substantial degree of manual work that needs to be done to track GPU resources.

Given the 64-bit address space, it would be an interesting approach. Some Australian university tried it a couple decades ago – a unified mapping that, presumably, only made select regions visible to a given process. They said it offered accelerated task launch, and obviously would simplify inter-process communication. I would imagine it could also provide for fixed address space for code (this particular app always loads at 0xFFB7400851331080). Their idea was to have files loading into the same spot every time, so that literal pointers could be used in the file, about which I would be skeptical, but the basic idea seems sound.

 

[Cmaier]

Given the 64-bit address space, it would be an interesting approach. Some Australian university tried it a couple decades ago – a unified mapping that, presumably, only made select regions visible to a given process. They said it offered accelerated task launch, and obviously would simplify inter-process communication. I would imagine it could also provide for fixed address space for code (this particular app always loads at 0xFFB7400851331080). Their idea was to have files loading into the same spot every time, so that literal pointers could be used in the file, about which I would be skeptical, but the basic idea seems sound.

Sort of seems like a security problem to load things into the same spot every time.

 

[Yoused]

Sort of seems like a security problem to load things into the same spot every time.

I can see your point. But if an app's root points to a mapping that does not include the region in question, it should not be a problem. I would be inclined to crash out any process that tries to access any othermise-mapped region that is not in its own mapping, so that iterative timing tricks are bound to fail.

Also, although the code area would be statically mapped, I would set the OS to initially randomize it (no two installations are the same) and heavily lard it with guard pages (again, making any access to a guard page fatal, to prevent scans). Nearly all code would be in all the maps (R/O, all other regions NX), but casually examining it to find stuff you want to play with would kill your program.

Most apps, AFAIK, do not rely on MMIO file data, and if more did, flat mapping would probably tend to become problematic. Fixed position file data does not sound like a good idea. Handling data offsets instead of literal pointers costs basically nothing.

 

[dada_dave]

That is a server/cluster config and I doubt Apple will build something that big. Are you just referring to the structure of having to connected CPUs? I think Apple might do something like that but is more likely to continue down the Ultra fusion route and try to have everything appear to be 1 chip with no NUMA.

@exoticspice1 Btw I was slightly wrong here. This is an Apple style ultra fusion chiplet design. I got confused by the reuse of the nvlink name but this has a similar (slightly higher) bandwidth as ultra fusion and seems to be similar upon closer reading. So this is the overall style in which Apple probably would have/will make an “Mx extreme” chip but bigger as it is meant for data center/clusters. I doubt Apple will ever create something quite that big as I they tend to be consumer focused.

 

[Colstan]

Using fell magic, dark arts, and a scrying pool, I've been able to divine the answer concerning the expandability of the Apple Silicon Mac Pro, if Apple decides to no longer support third-party graphics cards, which currently seems to be the case. I have finally found the answer.

Exhibit "A" are the Sonnet Echo I and Echo III expansion modules. If you go down to the section that says "Versatile PCIe Card Expansion" it will have a link to a PDF for supported PCIe cards. So that you don't have to wade through that page, here is the direct link for the relevant PDF. Take a gander at "M1/M2 Mac Compatible Driver". There are currently 60+ PCIe cards that are supported by Apple Silicon. None of those are GPUs.

So, even if Apple doesn't provide graphics card support for the next Mac Pro, there is certainly plenty reason to have PCIe slots.

 

[dada_dave]

Using fell magic, dark arts, and a scrying pool, I've been able to divine the answer concerning the expandability of the Apple Silicon Mac Pro, if Apple decides to no longer support third-party graphics cards, which currently seems to be the case. I have finally found the answer.

Exhibit "A" are the Sonnet Echo I and Echo III expansion modules. If you go down to the section that says "Versatile PCIe Card Expansion" it will have a link to a PDF for supported PCIe cards. So that you don't have to wade through that page, here is the direct link for the relevant PDF. Take a gander at "M1/M2 Mac Compatible Driver". There are currently 60+ PCIe cards that are supported by Apple Silicon. None of those are GPUs.

So, even if Apple doesn't provide graphics card support for the next Mac Pro, there is certainly plenty reason to have PCIe slots.

I think the issue is less should the Mac Pro have PCIe slots and more - is an Ultra Mac Pro with internal PCIe slots going to be that attractive over just getting an Ultra Mac Studio and using thunderbolt. For some? Yes absolutely. But enough to justify an entire product tier? Hmmm … that’s where it becomes a problem if the Extreme doesn’t exist at all and an internal PCIe slot isn’t technically necessary as a thunderbolt connection can accomplish the same task. For a dGPU you can take a tangible hit over thunderbolt so an internal slot can make a difference, but for most other cards you won’t notice the difference. As @Citysnaps would argue an internal slot can be far more convenient, but even so, to my mind there has to be something more, to really set the Pro apart and make it worth existing.

Edit: Apple has stated each product in the lineup has to justify its existence: what does this product bring to the table that products above and especially below it in the stack can’t accomplish. If the answer isn’t convincing it needs to be cut. Now to be clear: I am not calling for cutting the Mac Pro. The dang thing hasn’t even been released yet so let’s see it’s capabilities first. But a huge CPU, huge GPU, and expandable RAM are all major reasons the Pro existed in the first place and the rumors and logical deduction point to at least the last being moot with an ASi Pro and possibly even the first two as well - plausibly no dGPUs available and if no extreme, then no super high core count CPU or GPU at all.

So people are theorizing what could could Apple do that wouldn’t violate their design principles AND still give a reason for the product to exist. And in my mind an internal PCIe slot by itself won’t cut it (full disclosure it might for me if Hector is wrong and CUDA will function on Asahi Linux with an attached Nvidia dGPU but that’s a really small corner case).

 

[theorist9]

I think the issue is less should the Mac Pro have PCIe slots and more - is an Ultra Mac Pro with internal PCIe slots going to be that attractive over just getting an Ultra Mac Studio and using thunderbolt. For some? Yes absolutely. But enough to justify an entire product tier? Hmmm … that’s where it becomes a problem if the Extreme doesn’t exist at all and an internal PCIe slot isn’t technically necessary as a thunderbolt connection can accomplish the same task. For a dGPU you can take a tangible hit over thunderbolt so an internal slot can make a difference, but for most other cards you won’t notice the difference. As @Citysnaps would argue an internal slot can be far more convenient, but even so, to my mind there has to be something more, to really set the Pro apart and make it worth existing.

It seems there are three ways Apple could distinguish an M2 Ultra MP from a future M2 Ultra Studio without having to develop an entirely new chip architecture for the MP, and while making use of the existing MP case:

1) PCIe slots.
2) Use the MP's beefy cooling to boost CPU and GPU clocks.
3) Upgradeable internal storage.

[I was going to add LPDDR5X RAM, but that will probably be in the M2 Studio.]

As to whether these will be enough, I think it comes down to pricing.

 

[Colstan]

I think the issue is less should the Mac Pro have PCIe slots and more - is an Ultra Mac Pro with internal PCIe slots going to be that attractive over just getting an Ultra Mac Studio and using thunderbolt.

Concerning price, take a $4,999 Mac Studio Ultra and combine it with a $1,649 Sonnet xMac Echo III. That gets you to nearly $6,700, well within Apple's entry-level price range.

But enough to justify an entire product tier?

That's a different question. I was simply answering whether there is a need for PCIe cards other than for graphics. Clearly there is, otherwise Sonnet wouldn't be selling such a device.

So people are theorizing what could could Apple do that wouldn’t violate their design principles AND still give a reason for the product to exist.

I agree with @theorist9, and I think higher clocks will be the biggest selling point thanks to beefy cooling and more juice.

 

[dada_dave]

It seems there are three ways Apple could distinguish an M2 Ultra MP from a future M2 Ultra Studio without having to develop an entirely new chip architecture for the MP, and while making use of the existing MP case:

1) PCIe slots.
2) Use the MP's beefy cooling to boost CPU and GPU clocks.
3) Upgradeable internal storage.

[I was going to add LPDDR5X RAM, but that will probably be in the M2 Studio.]

As to whether these will be enough, I think it comes down to pricing.

Concerning price, take a $4,999 Mac Studio Ultra and combine it with a $1,649 Sonnet xMac Echo III. That gets you to nearly $6,700, well within Apple's entry-level price range.
I agree with @theorist9, and I think higher clocks will be the biggest selling point thanks to beefy cooling and more juice.

Supposedly it’ll be a new smaller pro case with a smaller number of full sized PCIe slots. That’s the latest rumor I remember anyway. The Ultra Studio already appears to have a damn good cooling system - it’s not clear that it’s operating anywhere near its thermal capacity. And with nothing else internal storage/PCIe slots by themselves are too thin in my mind.

More than just pricing, as I wrote above in my edit: Apple has stated each product in the lineup has to justify its existence: what does this product bring to the table that products above and especially below it in the stack can’t accomplish. If the answer isn’t convincing it needs to be cut.

That's a different question. I was simply answering whether there is a need for PCIe cards other than for graphics. Clearly there is, otherwise Sonnet wouldn't be selling such a device.

No one said otherwise. The question posed in this thread by myself and others is just having an internal PCIe slot worth an entire product tier with no or few other other distinguishing features. Let me turn it around: the fact that Sonnet sells such a widget and for most devices outside of dGPUs that will be enough is proof that internal PCIe isn’t necessary. Again I’m not arguing to kill the Pro but I want to see something to justify its existence that’s meaty that you can really point to and say “yes! That’s why this exists!”

Edit: let me rephrase, internal expansion may be a necessary feature of a Pro workstation, but it isn’t sufficient especially if there’s nothing that internal expansion is itself absolutely necessary for and much of the other expansiveness is gone. People will buy an M2 Ultra Pro - hell if I have the funds I might!, but if there’s no Extreme chip and nothing really juicy to plug into that remaining full sized PCIe slot and no socketable SOC or expandable RAM, then that’s a tough sell to the wider Pro market. That’s going to be underwhelming.

Btw I think Apple will come out with something here. I don’t countenance all these rumors (though some of them make sense). So I’m not here to cast doom and gloom that the Mac Pro will be underwhelming. But I do agree that if all the rumors are true and there’s little else beyond them that would be an unfortunate introduction for the ASi Mac Pro in what should be a marquee halo product.

 

[Citysnaps]

Regarding PCIe slots - internal vs external card cage via TB...

I think it would be valuable having professional music producers/sound designers/composers/score editors/etc. weigh in on that. At least in creative audio space, there are other uses, too.

Neil Parfitt (above) seems to prefer having his stuff all in a MacPro.

Hopefully Apple has him on-board as a consultant as he noted a bunch of irritating things Apple can improve on with respect to the current rack mount MacPro.

 

[dada_dave]

Regarding PCIe slots - internal vs external card cage via TB...

I think it would be valuable having professional music producers/sound designers/composers/score editors/etc. weigh in on that. At least in creative audio space, there are other uses, too.

Neil Parfitt (above) seems to prefer having his stuff all in a MacPro.

Hopefully Apple has him on-board as a consultant as he noted a bunch of irritating things Apple can improve on with respect to the current rack mount MacPro.

Again my position is not that a Mac Pro shouldn’t have internal expansion - it absolutely should! - though the current rumors are it will be smaller with less. But rather my argument is that it has to have more available to it than just a PCIe and M.2 slot. There has to be a true workstation level chip or some sort of expandability that really pushes the PCIe slot or a socket for the SOC or something.

 

[Citysnaps]

Again my position is not that a Mac Pro shouldn’t have internal expansion - it absolutely should! - though the current rumors are it will be smaller with less. But rather my argument is that it has to have more available to it than just a PCIe and M.2 slot. There has to be a true workstation level chip or some sort of expandability that really pushes the PCIe slot or a socket for the SOC or something.

For sure on more than a single slot. One slot will not address the bulk of the market, imo.

Professional sound editor Neil Parfitt has his rack mount MacPro filled up.

His unboxing vid I posted up above (not the one previously posted) goes into his setup in a lot of detail. As I mentioned earlier, he's the real deal. And imo, worth listening to.

 

[Colstan]

Supposedly it’ll be a new smaller pro case with a smaller number of full sized PCIe slots.

Current rumors suggest recycling of the the same case as the 2019 model, reduction to six PCIe slots.

No one said otherwise.

I was imprecise with my language, allow me to correct that. I have heard the argument that the primary justification for the Mac Pro is graphics cards compatibility. I believe that the Sonnet device and 60+ PCIe card list suggests that the non-GPU market may be sufficient. It's not going to make the enthusiast crowd happy, but it may be enough.

Again I’m not arguing to kill the Pro but I want to see something to justify its existence that’s meaty that you can really point to and say “yes! That’s why this exists!”

To be completely honest (as opposed to dishonest), I don't know why the Mac Pro wasn't canceled years ago. My suspicion, and this is nothing more than a suspicion, is that the iMac Pro was supposed to replace the Mac Pro. I think the 2019 Mac Pro was a result of the backlash and scramble before all pros left for other platforms. By now, I would be surprised if Apple hasn't lost even more of those customers.

Even if I am wrong, I think it's fairly obvious that Apple has spent the past ten years trying to figure out what it wants to do with the thing. They probably keep it around as a halo product, but I'm with @Cmaier, in that the Mac Pro is precariously sitting under a Sword of Damocles.

 

[dada_dave]

Current rumors suggest recycling of the the same case as the 2019 model, reduction to six PCIe slots.

I think one full sized though? Or am I misremembering?

I was imprecise with my language, allow me to correct that. I have heard the argument that the primary justification for the Mac Pro is graphics cards compatibility. I believe that the Sonnet device and 60+ PCIe card list suggests that the non-GPU market may be sufficient. It's not going to make the enthusiast crowd happy, but it may be enough.

I think it was more well the rumor is the Extreme is dead, maybe it’ll have a dGPU as unlikely as it seems - that would really push the need for internal PCIe, no that’s dead too, maybe it’s expandable RAM, no that’s dead too.

To be completely honest (as opposed to dishonest), I don't know why the Mac Pro wasn't canceled years ago. My suspicion, and this is nothing more than a suspicion, is that the iMac Pro was supposed to replace the Mac Pro. I think the 2019 Mac Pro was a result of the backlash and scramble before all pros left for other platforms. By now, I would be surprised if Apple hasn't lost even more of those customers.

Even if I am wrong, I think it's fairly obvious that Apple has spent the past ten years trying to figure out what it wants to do with the thing. They probably keep it around as a halo product, but I'm with @Cmaier, in that the Mac Pro is precariously sitting under a Sword of Damocles.

I think iMac Pro was a hold over until the new Mac Pro came out, but if Apple doesn’t come out with an extreme chip with really high core counts, if the ultra remains the top of the line, an entire product line with a single low volume processor with an even more niche utility. Then agreed it really would be under the proverbial sword.

But a high end workstation running macOS where the cpu and GPU core count can go hog wild? With Apple’s powerful but efficient processors? I can see THAT being something that would turn heads. Yes it’s still niche in some respects, but it’s a big niche and one that could really grow depending on its capabilities: animators, artists, scientists, etc … personally I think the notion that Apple will never introduce an extreme is bunk. They will. But it will be unfortunate if it doesn’t come out with the Mac Pro.

 

[Citysnaps]

but it’s a big niche and one that could really grow depending on its capabilities: animators, artists, scientists, etc …

Totally agree. And I believe that that the market is already using them; in all of the categories you mentioned.

I just hope Apple is talking to real users; rather than internal EEs/CSEEs who indeed may be very smart and capable, may not understand what real users would really like; and thus make poor strategic decisions.

 

[Cmaier]

Current rumors suggest recycling of the the same case as the 2019 model, reduction to six PCIe slots.

I was imprecise with my language, allow me to correct that. I have heard the argument that the primary justification for the Mac Pro is graphics cards compatibility. I believe that the Sonnet device and 60+ PCIe card list suggests that the non-GPU market may be sufficient. It's not going to make the enthusiast crowd happy, but it may be enough.

To be completely honest (as opposed to dishonest), I don't know why the Mac Pro wasn't canceled years ago. My suspicion, and this is nothing more than a suspicion, is that the iMac Pro was supposed to replace the Mac Pro. I think the 2019 Mac Pro was a result of the backlash and scramble before all pros left for other platforms. By now, I would be surprised if Apple hasn't lost even more of those customers.

Even if I am wrong, I think it's fairly obvious that Apple has spent the past ten years trying to figure out what it wants to do with the thing. They probably keep it around as a halo product, but I'm with @Cmaier, in that the Mac Pro is precariously sitting under a Sword of Damocles.

There are doubtless engineers at Apple who really like using mac pros, which is probably part of what keeps them coming. As long as Apple doesn’t lose money on them, they’re likely safe. But if they are going to get into the habit of making CPUs/SoCs just for Mac Pro, there’s a good chance they could end up losing money on them.

 

[Citysnaps]

Writing the above recent posts makes me wonder how Apple's product development is organized with respect to engineering and marketing resources.