The original post and none of the subsequent posts mention what is almost certainly the biggest stumbling block to supporting non-Apple GPU hardware: drivers.
Drivers have always been the biggest issue with Apple GPU support, and it has always been a hot potato tossed back and forth between Apple's OS group and the 3rd party HW vendor (including Intel for the integrated GPUs). GPU drivers are terribly complex things, and Apple can't/doesn't use the drivers written by AMD/Intel/nVidia... and those vendors aren't likely to put much effort into writing drivers for macOS even if Apple were to start shipping their GPUs in Apple products. They never did before, the market is too small. So will Apple write drivers for any 3rd party devices? Their current direction suggests that the answer is a resounding "no", but that's not definitive and could change. They still have drivers that work on the Intel chip based Macs, and porting to Aarch64 may not be terribly difficult. Keeping up with the moving target that is the latest AMD GPUs though, is a lot of work. On top of supporting Apple's own GPU designs.
The Apple Silicon hardware is almost certainly hardware compatible with most GPUs from other vendors, thanks to PCI-e / Thunderbolt being standardized in its various flavours. So you can physical install any of the devices, but you need drivers to make it interoperate with macOS and macOS needs to continue to expose the functionality required to do that (which conceivably it may not on Apple Silicon since the macOS team may be taking advantage of detailed knowledge of the hardware).
Which is why I think anything Apple does as far as third-party GPUs will be short term, I think Apple is working towards not being dependent upon outside, third-party companies when it comes to their graphics systems, but to do so, takes long-term effort time and the will to do it.
What’s talked about as covered by these patent applications is old hat when applied in other contexts, so this reads much like Apple attempting to cover the concepts “with a GPU or GPUs” to protect themselves as much as possible.
TBD if these patents are considered valid and enforceable, assuming they’re ever actually implemented in the market by anyone.
Think about how the M2 Pro relates to the M2 Max. It’s the same CPU, but the Pro has one GPU etc. and the Max has two GPUs etc. That’s different from the Ultra (assuming the M2 follows the M1 approach), which is just a dual-Max design. It’s not the same type of difference as the Max versus the Pro, in terms of GPU etc. power relative to CPU power. That relationship doesn’t change in the Ultra.
If indeed the Extreme was a quad-Max design, then it would have also suffered from this limitation. Lots of CPU, but not enough GPU etc. But what if the key to the sort of exponential performance gains Apple is looking for is continuing in the same direction as the Pro-to-Max progression?
So the Ultra is for the Studio, and would be sufficient in a Mac Pro for some workloads, but the Extreme (as originally conceived) didn’t provide enough GPU etc. to do what they wanted to do with the Mac Pro. So they didn’t build it.
Apple can do this. Why? Because they aren’t limited to industry standards that all OEMs, big and small, use.
They don’t just take a standard from someone else, accept its limitations, and deal with it. They improve on those things where possible and even invent their own standards.
If Apple wants to use Apple Silicon as the base structure and add helper cards, they can definitely do that and it will work. It won’t scale the same as having it all one one SOC, but it will work and it will outperform everyone else.
A better option would be to introduce desktop specific SOCs without mobile limitations or power efficiency concerns. The MC Peo is THE perfect candidate for such a beast. Take the cheese grater case, the massive fan setup snd give it something to match. Give it all performance cores and lots of them, crank up the GHZ, pop the new ray tracing GPU cores in there also lots of them with jacked up clock speeds and let it rip.
Aside from that, Apple could add an entirely new spin on the modular approach and instead of adding generic PCI-E slots, throw that old stuff away and build out a light speed socketed “fabric” that allows multiple Apple Silicon SOCs to be plugged in as the user requires. Need more RAM? It happens to come with more CPU and GPU power also and vice versa.
Apple has not usually just done what everyone else thinks is possible. iTunes, the iPod, iPhone, Mac Pro, Apple Silicon, Force Touch trackpads, etc. are all examples of Apple doing things way differently and better than everyone else, paving the way for a better computing landscape.
All it takes os for Apple to want to do this.
Sure Apple COULD do the PCI thing. But that’s the lazy and inefficient way.
I could see either a desktop specific SOC and/or a system fabric that connect multiple Apple Silicon SOCs together as needed. I can also see each SOC module coming with its own cooling system. Apple could also have modular power supplies to handle energy needs per configuration and sell those after purchase as well. For example, if you buy an M3 Extreme Mac Pro, it comes with a suitable power supply. But if you add 3 more M3 Extremes later, you could buy the necessary power supply from Apple, replace the old power supply module, and presto, supercomputer.
Apple has an opportunity to reinvent the desktop computer here, redefine modularity in meaningful ways, offer a good deal, while also opening up profit margin growth after the initial sale, and leave everyone else in the dust. This would also add continual shine to Apple Silicon as the ultimate platform.
This is the Mac Pro - the absolute pinnacle of Apple performance snd represents the best computer Apple could possibly make. It’s their time to shine. Hopefully that’s what they do. Just don’t rush it. Even if it takes another year, get it right and set the tone for the next decade to 20 years.
I was thinking the same thing regarding the PCI bus. Apple could create a "new bus" (see what I did there?)...
...then Apple could publish specs so that 3rd party solutions could be made available for things that Apple is less interested in producing due to low-volume (but hopefully, enough volume to be worthwhile). In this case, Apple would not be dependent on 3rd party hardware vendors, but would help promote an ecosystem somewhat similar to how an app ecosystem works, thereby increasing the value of the platform. (Hardware wouldn't need to be confined to GPUs either). I could see Blackmagic being interested. And who knows, AMD and as crazy as this sounds, even Nvidia could be interested in being part of a new "supercomputer" platform.
Now there is one thing that Programmer on comment #20 pointed out, and that is the driver problem. I don't know what the answer to this is, but there are a whole lot of smart people out there, so maybe this could be solved. Couldn't Apple create some sort of "universal driver(s)" that 3rd party vendors could take advantage of?
No drivers needed. Apple isn’t looking to give AMD and Nvidia business any more than they are looking to sell Apple Hardware.
Offering equipment from the other guys would also be communicating that they really couldn’t compete at the top level. And that would be wrong.
The Mac Pro will be all Apple Silicon. It will likely be a multiple-SOC setup inside a case that may actually be even larger than the current housing.
I have a feeling that Apple doesn’t just want people to be able to purchase SOC modules as that could open the door to their hardware being experimented on in ways Apple wouldn’t deem desirable.
What I can see is a special Apple Store program where the customer ships/brings their Mac Pro in and Apple does the install - and solders the SOC in. This would be a pain if the Mac Pro sold as many units as an iPhone, etc. but as a niche product, it wouldn’t cause too much overhead.
While this may take the tinker toy fun out of upgrades, it ensures a proper install while also mitigating too much tampering with Apples SOCs.
Maybe I’m misunderstanding, but this sounds like software would not need to be aware of different latencies for different GPUs — it sounds like Apple’s stack would handle that transparently. If so, then these patents seriously change my mind about the potential for discrete GPUs in the Mac Pro (and perhaps other macs too). One of the big benefits apple has touted is the shared memory space and how developers need not worry about having a distant GPU. I thought a discrete GPU would break that model, but this sounds like maybe not.
I would love to be totally wrong on this, but after thoroughly reading each patent application it looks like these are the patents for how the Ultra chip (two Max chip linked together through infinity fabric) processes graphic task in tandem, and the infinity fabric link which Apple uses to tie both MAX chips together, and has nothing to do with future multiple dGPU support by Apple.
Further evidence of this is that the patent file date is for August 2021 and the Ultra chip debuted in March 2022.
This doesn’t mean that Apple isn’t working on dGPUs though. And we may very well still see dGPU support come to the new AS MacPros. I just think these patent applications cover the Apple Silicon Ultra’s pair of GPUs talking to each other in tandem.
Completely agree. When the Ultra debuted people wondered how Apple was treating each "GPU". Were they seen by the kernel as 2 discrete GPUs, or were they tied together as 1 in hardware? This patent definitely alludes to this. I think the interpretation is thrown off by the use of "slot". It's easy to make the leap to "PCI slot" without any further context.
Several facts...
1. Apple is already adept at supporting multiple GPUs, be it discrete or integrated or both. 2. M-series already supports PCI-e, can't have Thunderbolt without it. (M-series Macs already support PCI breakout boxes.) 3. PCI-e is just a data bus (a peripheral component interface). It doesn't matter what you plug in on the other end as long you can interpret and manipulate the data that is sent and received (via drivers). 4. macOS is extremely capable at process dispatching. It is not at all dependent on integrated processing units of any kind.
Opinion...
1. Apple is not supporting dGPUs currently because at this early stage they're more interested in forcing all developers to adopt and optimize code for Apple's GPUs. 2. Performance wise, Apple's GPUs aren't quite where they were hoping for - yet. And until they can approach the level of performance that most 3rd party GPUs are, they're holding back support.
I would love to be totally wrong on this, but after thoroughly reading each patent application it looks like these are the patents for how the Ultra chip (two Max chip linked together through infinity fabric) processes graphic task in tandem, and the infinity fabric link which Apple uses to tie both MAX chips together, and has nothing to do with future multiple dGPU support by Apple.
Further evidence of this is that the patent file date is for August 2021 and the Ultra chip debuted in March 2022.
This doesn’t mean that Apple isn’t working on dGPUs though. And we may very well still see dGPU support come to the new AS MacPros. I just think these patent applications cover the Apple Silicon Ultra’s pair of GPUs talking to each other in tandem.
Completely agree. When the Ultra debuted people wondered how Apple was treating each "GPU". Were they seen by the kernel as 2 discrete GPUs, or were they tied together as 1 in hardware? This patent definitely alludes to this. I think the interpretation is thrown off by the use of "slot". It's easy to make the leap to "PCI slot" without any further context.
Several facts...
1. Apple is already adept at supporting multiple GPUs, be it discrete or integrated or both. 2. M-series already supports PCI-e, can't have Thunderbolt without it. (M-series Macs already support PCI breakout boxes.) 3. PCI-e is just a data bus (a peripheral component interface). It doesn't matter what you plug in on the other end as long you can interpret and manipulate the data that is sent and received (via drivers). 4. macOS is extremely capable at process dispatching. It is not at all dependent on integrated processing units of any kind.
Opinion...
1. Apple is not supporting dGPUs currently because at this early stage they're more interested in forcing all developers to adopt and optimize code for Apple's GPUs. 2. Performance wise, Apple's GPUs aren't quite where they were hoping for - yet. And until they can approach the level of performance that most 3rd party GPUs are, they're holding back support.
What I think is happening there is that Apple is taking the Max Ultra “Fabric” concept and applying that to an entire “logic board.” No PCI-E, no legacy stuff of any kind (outside of I/o ports). An entirely new way to connect performance modules in a lightning fast way, via a series of hardware controllers, that the OS sees as one resource and scales similarly to having all of these modules (full Mx SOCs quite possibly) actually on one die.
This would blow the minds of every reviewer, critic, nerd, power user, and cynic.
And it looks like it’s actually happening.
Been waiting for an iMac 30+” but if the Mac Pro becomes what I think it’s going to, I’ll be owning my first Mac Pro by early 2024.
I basically said what 9secondkox2 said so I’m deleting my post
Let’s hope we see the light of day of this at WWDC this year, if not sooner! But I think WWDC will be more appropriate so they can get full press coverage, as they will likely have an in-person keynote, or at the very least invite the press into a room where these machines will be on demo after, likely with the VR headset.
I would love to be totally wrong on this, but after thoroughly reading each patent application it looks like these are the patents for how the Ultra chip (two Max chip linked together through infinity fabric) processes graphic task in tandem, and the infinity fabric link which Apple uses to tie both MAX chips together, and has nothing to do with future multiple dGPU support by Apple.
Further evidence of this is that the patent file date is for August 2021 and the Ultra chip debuted in March 2022.
This doesn’t mean that Apple isn’t working on dGPUs though. And we may very well still see dGPU support come to the new AS MacPros. I just think these patent applications cover the Apple Silicon Ultra’s pair of GPUs talking to each other in tandem.
The UltraFusion silicon bridge that interconnects the M1 Ultra is not "infinity fabric," which is an AMD technology. UltraFusion uses the TSMC InFO_LI (or InFO_LSI) packaging method. It's a passive, "local" silicon interconnect, something like Intel's EMIB (embedded die interconnect bridge) packaging. Dylan Patel says "it can evolve into being active (transistors and various IP) in the future." [https://www.semianalysis.com/p/advanced-packaging-part-2-review]
So these patents could be aimed at the next (or a future) generation of UltraFusion. Note that Infinity Fabric (now in its fourth generation and marketed as Infinity Architecture) isn't a silicon interposer like UltraFusion. Using silicon to connect chip-to-chip is very expensive, which is why Infinity Fabric was created.
So I think it's possible that Apple, unlike AMD (which as a "merchant silicon vendor" has to have the ability to lower prices in the face of competition), can work with TSMC to link additional GPU cores to M2 Ultra in a Mac Pro using silicon inside of an end product. For AMD, the product is the chip. For Apple, the product is this M1 Mac mini I'm typing this on, and the iPhone I first noticed this article on.
That said, AMD's aims with the current generation of Infinity Architecture are probably instructive for Apple. The most extreme configuration it supports are two CPU chiplets and eight GPU chiplets, a ratio of 1:4 (1 CPU to 4 GPU), all interconnected using the architecture. The M1 Max has a ratio of 1:2 (1 CPU to 2 GPU). The M2 Ultra doesn't improve on this ratio, with two CPU and four GPU. Apple has an advantage because it uses a silicon interconnect, but that's not enough. So I think (in my layman's ignorance) that might have been the fundamental problem—a quad M1 Extreme doesn't improve on the CPU-to-GPU ratio. They need a ratio of 1:4 to really compete with high-end AMD configurations. Not to mention Intel and Nvidia.
Finally, one last observation: the M1/M2 Pro is thought to be cut down from the M1/M2 Max. I don't know if this is actually true, but that's what I've read. So they cut one of the two GPU units off the Max to make the Pro. When they do that, there is a GPU unit left over. Some of them may have flaws that make them unusable, but demand for the M2 Pro (especially now with the M2 Pro Mac mini) is high enough that some percentage of them must also be flawless. What to do with them? Put four of them (for M2 Ultra), or eight of them (for M2 Extreme), into a GPU extension that uses the next generation of UltraFusion silicon interconnect (i.e., not "local") to add GPU power, to reach that 1:4 ratio.
I would love to be totally wrong on this, but after thoroughly reading each patent application it looks like these are the patents for how the Ultra chip (two Max chip linked together through infinity fabric) processes graphic task in tandem, and the infinity fabric link which Apple uses to tie both MAX chips together, and has nothing to do with future multiple dGPU support by Apple.
Further evidence of this is that the patent file date is for August 2021 and the Ultra chip debuted in March 2022.
This doesn’t mean that Apple isn’t working on dGPUs though. And we may very well still see dGPU support come to the new AS MacPros. I just think these patent applications cover the Apple Silicon Ultra’s pair of GPUs talking to each other in tandem.
Completely agree. When the Ultra debuted people wondered how Apple was treating each "GPU". Were they seen by the kernel as 2 discrete GPUs, or were they tied together as 1 in hardware? This patent definitely alludes to this. I think the interpretation is thrown off by the use of "slot". It's easy to make the leap to "PCI slot" without any further context.
Several facts...
1. Apple is already adept at supporting multiple GPUs, be it discrete or integrated or both. 2. M-series already supports PCI-e, can't have Thunderbolt without it. (M-series Macs already support PCI breakout boxes.) 3. PCI-e is just a data bus (a peripheral component interface). It doesn't matter what you plug in on the other end as long you can interpret and manipulate the data that is sent and received (via drivers). 4. macOS is extremely capable at process dispatching. It is not at all dependent on integrated processing units of any kind.
Opinion...
1. Apple is not supporting dGPUs currently because at this early stage they're more interested in forcing all developers to adopt and optimize code for Apple's GPUs. 2. Performance wise, Apple's GPUs aren't quite where they were hoping for - yet. And until they can approach the level of performance that most 3rd party GPUs are, they're holding back support.
What I think is happening there is that Apple is taking the Max Ultra “Fabric” concept and applying that to an entire “logic board.” No PCI-E, no legacy stuff of any kind (outside of I/o ports). An entirely new way to connect performance modules in a lightning fast way, via a series of hardware controllers, that the OS sees as one resource and scales similarly to having all of these modules (full Mx SOCs quite possibly) actually on one die.
This would blow the minds of every reviewer, critic, nerd, power user, and cynic.
And it looks like it’s actually happening.
Been waiting for an iMac 30+” but if the Mac Pro becomes what I think it’s going to, I’ll be owning my first Mac Pro by early 2024.
When was the last time Apple went alone on such a low level system and it paid off? It has been awhile.
If they are going something new then it would be new thing the industry is abuzz with better still if they can contribute too. Apple have said they will only do their own where they can do better for Mac Pro even a MacStudio they can't do better they need to open the door.
When was the last time Apple went alone on such a low level system and it paid off? It has been awhile.
If they are going something new then it would be new thing the industry is abuzz with better still if they can contribute too. Apple have said they will only do their own where they can do better for Mac Pro even a MacStudio they can't do better they need to open the door.
Well... Apple Silicon itself is such a "low level system". On one hand, I agree that I don't think we will see Apple create an "all singing, all dancing" board level interconnect. On the other hand, that isn't what is needed here. They did the edge connect between 2 M1 Max chips to produce the M1 Ultra, not to mention the on-SoC network in their silicon itself... so there is some indication that they are willing to work in this space. Realistically, they're going to have to do something because chiplet-based SoCs are where the industry seems to be going. At the chiplet in-package level, I think we could see Apple do something, but at the board level, I doubt it. They'd probably just use the latest PCI-E, or perhaps even license a fabric from someone. If both sides of the connection aren't Apple Silicon, its going to be an industry standard interconnect (PCI-E). If both sides are Apple Silicon, it looks like that will only be an in-package interconnect. An Apple Silicon to Apple Silicon interconnect at the board level would be a surprise, and would (almost certainly) be unique to Mac Pro... and thus not worth them investing enormously into a hugely expensive R&D effort. So if the Mac Pro does contain multiple AS packages (and I'm doubting it), I would expect an off-the-shelf interconnect (most likely PCI-E as they already have it in their hardware, but they could also license something else). In the short term, therefore, I expect M2 Ultra. In the M3 timeframe, bigger packages. Beyond that, its too hard to predict.
Comments
Which is why I think anything Apple does as far as third-party GPUs will be short term, I think Apple is working towards not being dependent upon outside, third-party companies when it comes to their graphics systems, but to do so, takes long-term effort time and the will to do it.
If indeed the Extreme was a quad-Max design, then it would have also suffered from this limitation. Lots of CPU, but not enough GPU etc. But what if the key to the sort of exponential performance gains Apple is looking for is continuing in the same direction as the Pro-to-Max progression?
...then Apple could publish specs so that 3rd party solutions could be made available for things that Apple is less interested in producing due to low-volume (but hopefully, enough volume to be worthwhile). In this case, Apple would not be dependent on 3rd party hardware vendors, but would help promote an ecosystem somewhat similar to how an app ecosystem works, thereby increasing the value of the platform. (Hardware wouldn't need to be confined to GPUs either). I could see Blackmagic being interested. And who knows, AMD and as crazy as this sounds, even Nvidia could be interested in being part of a new "supercomputer" platform.
Now there is one thing that Programmer on comment #20 pointed out, and that is the driver problem. I don't know what the answer to this is, but there are a whole lot of smart people out there, so maybe this could be solved. Couldn't Apple create some sort of "universal driver(s)" that 3rd party vendors could take advantage of?
Completely agree. When the Ultra debuted people wondered how Apple was treating each "GPU". Were they seen by the kernel as 2 discrete GPUs, or were they tied together as 1 in hardware? This patent definitely alludes to this. I think the interpretation is thrown off by the use of "slot". It's easy to make the leap to "PCI slot" without any further context.
Several facts...
1. Apple is already adept at supporting multiple GPUs, be it discrete or integrated or both.
2. M-series already supports PCI-e, can't have Thunderbolt without it. (M-series Macs already support PCI breakout boxes.)
3. PCI-e is just a data bus (a peripheral component interface). It doesn't matter what you plug in on the other end as long you can interpret and manipulate the data that is sent and received (via drivers).
4. macOS is extremely capable at process dispatching. It is not at all dependent on integrated processing units of any kind.
Opinion...
1. Apple is not supporting dGPUs currently because at this early stage they're more interested in forcing all developers to adopt and optimize code for Apple's GPUs.
2. Performance wise, Apple's GPUs aren't quite where they were hoping for - yet. And until they can approach the level of performance that most 3rd party GPUs are, they're holding back support.
Let’s hope we see the light of day of this at WWDC this year, if not sooner! But I think WWDC will be more appropriate so they can get full press coverage, as they will likely have an in-person keynote, or at the very least invite the press into a room where these machines will be on demo after, likely with the VR headset.
So these patents could be aimed at the next (or a future) generation of UltraFusion. Note that Infinity Fabric (now in its fourth generation and marketed as Infinity Architecture) isn't a silicon interposer like UltraFusion. Using silicon to connect chip-to-chip is very expensive, which is why Infinity Fabric was created.
So I think it's possible that Apple, unlike AMD (which as a "merchant silicon vendor" has to have the ability to lower prices in the face of competition), can work with TSMC to link additional GPU cores to M2 Ultra in a Mac Pro using silicon inside of an end product. For AMD, the product is the chip. For Apple, the product is this M1 Mac mini I'm typing this on, and the iPhone I first noticed this article on.
That said, AMD's aims with the current generation of Infinity Architecture are probably instructive for Apple. The most extreme configuration it supports are two CPU chiplets and eight GPU chiplets, a ratio of 1:4 (1 CPU to 4 GPU), all interconnected using the architecture. The M1 Max has a ratio of 1:2 (1 CPU to 2 GPU). The M2 Ultra doesn't improve on this ratio, with two CPU and four GPU. Apple has an advantage because it uses a silicon interconnect, but that's not enough. So I think (in my layman's ignorance) that might have been the fundamental problem—a quad M1 Extreme doesn't improve on the CPU-to-GPU ratio. They need a ratio of 1:4 to really compete with high-end AMD configurations. Not to mention Intel and Nvidia.
Finally, one last observation: the M1/M2 Pro is thought to be cut down from the M1/M2 Max. I don't know if this is actually true, but that's what I've read. So they cut one of the two GPU units off the Max to make the Pro. When they do that, there is a GPU unit left over. Some of them may have flaws that make them unusable, but demand for the M2 Pro (especially now with the M2 Pro Mac mini) is high enough that some percentage of them must also be flawless. What to do with them? Put four of them (for M2 Ultra), or eight of them (for M2 Extreme), into a GPU extension that uses the next generation of UltraFusion silicon interconnect (i.e., not "local") to add GPU power, to reach that 1:4 ratio.
It has been awhile.
If they are going something new then it would be new thing the industry is abuzz with better still if they can contribute too.
Apple have said they will only do their own where they can do better for Mac Pro even a MacStudio they can't do better they need to open the door.