See the thing that worries me most are the little things in apple's actions already. If they have behind the scenes test hardware for all their configurations already, and yet they couldn't manage to get to get Thunderbolt into their developers machine. Does that mean thunderbolt is gone for them, or they couldn't get it working? Either way is not a good thing considering how much they have leaned on Thunderbolt for pro work over the past years.
Couldn't manage to or had no need to?
No need? So thunderbolt developers have no need to make sure their hardware works? Or pro software that has hardware that attaches via thunderbolt? I would say for sure there is a need.
The A12Z Bionic is up to 8 GPU cores. The Most powerful and expensive GPUs have cores in the thousands. What would it take for Apple to create its own separate 500 core GPU SoC or maybe only a 100 core GPU with the ability to use several of them in a blade setup. There's nothing stopping Apple, other than patents, from making whatever they want to any way they want to. Look at the Mac Pro. It's a fantastic workstation.
It's all down to how they perform, rather than how the hardware is structured, some GPUs have more cores but run at lower clock speeds. Mobile GPUs (~7W) are around 1-2 TFLOPs just now. High-end GPUs (300W) are around 16TFLOPs and will be just over 20TFLOPs in the next generation. There's usually around 5-10x difference between high-end and low-end in a given generation of hardware. Apple should be able to put 4TFLOP GPUs in at the low-end (15W, Macbook Air). All it takes is 4 stacks or tiles of a GPU like that to be a high-end GPU (2 stacks/tiles for MBP = 8TFLOP, 4 stacks/tiles = 16TFLOPs iMac). The power profiles of those machines allow for more but commercially that's all they need in those and allows them to lower heat and fan noise.
The current fastest Mac Pro graphics option is 56TFLOPs as it uses quad 14TFLOP GPU, which costs over $10k. Reaching that level of performance in a single chip would be an achievement but isn't necessary because the pace of mobile chip development is faster. If they can do 16TFLOPs this year and 24-32TFLOPs in a couple of years, this kind of performance is enough for any professional workflow and they can supplement it with custom accelerators like they do with the AfterBurner and T2 chips.
I'd say the software differences will cause more issues. That support doc says they won't support OpenCL on the CPU:
It's good they will have OpenGL and OpenCL support so all that software can be ported over but OpenCL is only supported on the GPU. In theory that's ok because OpenCL usually runs faster there but often GPUs are less reliable for compute tasks. Intel demoed a 25x speedup using OpenCL on the CPU:
Hopefully Apple's GPU OpenCL implementation on their own hardware will be reliable but it hasn't been on AMD GPUs. If it's not reliable and developers have to fall back to unoptimized CPU code, that could result in some software running much more slowly. There's no way to know until people start testing this kind of software but it's an unusual omission leaving out OpenCL support from the CPU.
See the thing that worries me most are the little things in apple's actions already. If they have behind the scenes test hardware for all their configurations already, and yet they couldn't manage to get to get Thunderbolt into their developers machine. Does that mean thunderbolt is gone for them, or they couldn't get it working? Either way is not a good thing considering how much they have leaned on Thunderbolt for pro work over the past years.
Couldn't manage to or had no need to?
No need? So thunderbolt developers have no need to make sure their hardware works? Or pro software that has hardware that attaches via thunderbolt? I would say for sure there is a need.
I think for now the priority is software, and given there's no announced Mac hardware at this point they're not going to tip their hand with regard to I/O. For all we know they're working with those companies directly if they have specific needs to get them on board as soon as possible.
Or you could read it as that this is what the Apple Silicon processor is directly compatible with. It doesn't say they will not support third-party. I could see Apple using their own GPU for primary tasks but third-party for more beefy tasks. Do we really think that Apple could compete with a GPU like the AMD Radeon Pro Vega II for professionals?
FTA: "There is no indication in support documentation that Apple will discontinue support for AMD GPUs for Intel Macs in future versions of macOS, but the statement above may also suggest that there may yet be an avenue for third-party PCI-E GPU support going forward."
You have to admit though, that the chart is provocative. It could have showed an Apple GPU or a theirs party GPU, but it didn’t.
It is, yes.
Maybe! What I've been noticing is that there are a lot of people that read into Apple's documentation things that Apple has never said. The common one I've sen across the net is that people are quoting a statement by Apple that there will be i86 Macs for a long time to come. the problem is the sentence clearly says that Apple intends to support MACOS on i86 for a long time to come.
In this case we have people seemingly taking a statement that says nothing about discreet GPU's as a statement that Apple has no intention to use AMD GPU's in the future. It doesn't matter what Apple is doing in this case because it is a stretch to imply that AMD is out of the picture. In fact I just spent some time listening to a WWDC blurb on ray tracing support in Metal, this implies to me that they have plans that will likely require external hardware to achieve max performance. I say that because I don't see Apple having enough die space to be able to offer up a high performance ray tracing support. Now Apple could offer an external GPU or even a MCM but even the MCM could be supporting a third party chip. There are all sorts of options here.
I still think that if apple wants to offer a high performance Mac Pro, a third party GPU (or external compute chip) is a must for the next few years. I just see them needing a couple of years to offer up an ARM based chip that is similar in performance to Fijitsu's chip. Even then an accelerator chip can still make sense no matter how fast your CPU is. So instead of dropping AMD, I rather see them increasing their partnership with AMD. In this case that might mean new technology in Mac compute accelerations. Think a fabric connection to an external CDNA chip.
I'm finding it extraordinary difficult to believe that Apple could replace the existing Xeon-based Mac Pro (or even iMac Pro) with comparable performance & expandability in only two years. Low end, no problem at all.
It isn't that hard to believe, Fujitsu is already shipping a much faster ARM based chip. Expand ability is becoming a bit of a joke though, in the end I don't see a rush of users putting cards into their Mac Pro. The number of the people actually using those PCI-Express slots is extremely thin right now.
In any event you make a fundamental mistake here, Apple has been working on these chips for years now. I would imagine that they have already taped out more that a few models already.
Or you could read it as that this is what the Apple Silicon processor is directly compatible with. It doesn't say they will not support third-party. I could see Apple using their own GPU for primary tasks but third-party for more beefy tasks. Do we really think that Apple could compete with a GPU like the AMD Radeon Pro Vega II for professionals?
That chart is explicit. I take it at face value that Apple Silicon Macs will not support 3rd party GPU's. Anything otherwise is an assumption.
No it isn't. Please read again and again until you understand what it says or more importantly what it doesn't say.
See the thing that worries me most are the little things in apple's actions already. If they have behind the scenes test hardware for all their configurations already, and yet they couldn't manage to get to get Thunderbolt into their developers machine. Does that mean thunderbolt is gone for them, or they couldn't get it working? Either way is not a good thing considering how much they have leaned on Thunderbolt for pro work over the past years.
And the whole idea of no external graphics card, just everything on one the chip again means less chance of expansion and more chance of just having to keep upgrading the machines every few years. The thing that was so great about the old macpro was the ability upgrade the graphics card, and graphics on the chip are not going to be upgradeable.
And then there is the whole getting companies to support it, it is great that Adobe is working on it, but what about AVID. It took them this long to support Catalina and that is with AMD graphcis.
I really don't get why people stress out about the capability of the DTK. Honestly how much clearer does Apple have to state that the machine does not represent shipping hardware???? Really it blows the mind that people can't understand the simple statements that come from Apple.
As for the no external graphics card where in the hell does it say anything about that in the posted graphic??? Now I would fully expect Apple to not stay with the status quo on the Mac Pro when it comes to expansion, The fact is expansion has done nothing to accelerate Mac Pro sales. If any thing they have a dog of a machine on their hands.
The A12Z Bionic is up to 8 GPU cores. The Most powerful and expensive GPUs have cores in the thousands. What would it take for Apple to create its own separate 500 core GPU SoC or maybe only a 100 core GPU with the ability to use several of them in a blade setup. There's nothing stopping Apple, other than patents, from making whatever they want to any way they want to. Look at the Mac Pro. It's a fantastic workstation.
The Mac Pro is not a fantastic workstation,
Of course you don’t have to spend $14k on one. I purchased one for my studio for about $7500 including some aftermarket upgrades. I’ve owned many Macs since the time they used to be beige and the new MacPro is basically the absolute best Mac ever built. The performance is top notch (it’s not always just about benchmarks) and there are a slew of other reasons you failed to even include in your “analysis” like noise level(which is basically 0 ), top notch components, build quality and materials, overall design (the slip cover chassis is brilliant), expansion options and aesthetics.
Or you could read it as that this is what the Apple Silicon processor is directly compatible with. It doesn't say they will not support third-party. I could see Apple using their own GPU for primary tasks but third-party for more beefy tasks. Do we really think that Apple could compete with a GPU like the AMD Radeon Pro Vega II for professionals?
FTA: "There is no indication in support documentation that Apple will discontinue support for AMD GPUs for Intel Macs in future versions of macOS, but the statement above may also suggest that there may yet be an avenue for third-party PCI-E GPU support going forward."
You have to admit though, that the chart is provocative. It could have showed an Apple GPU or a theirs party GPU, but it didn’t.
It is, yes.
Maybe! What I've been noticing is that there are a lot of people that read into Apple's documentation things that Apple has never said. The common one I've sen across the net is that people are quoting a statement by Apple that there will be i86 Macs for a long time to come. the problem is the sentence clearly says that Apple intends to support MACOS on i86 for a long time to come.
In this case we have people seemingly taking a statement that says nothing about discreet GPU's as a statement that Apple has no intention to use AMD GPU's in the future. It doesn't matter what Apple is doing in this case because it is a stretch to imply that AMD is out of the picture. In fact I just spent some time listening to a WWDC blurb on ray tracing support in Metal, this implies to me that they have plans that will likely require external hardware to achieve max performance. I say that because I don't see Apple having enough die space to be able to offer up a high performance ray tracing support. Now Apple could offer an external GPU or even a MCM but even the MCM could be supporting a third party chip. There are all sorts of options here.
I still think that if apple wants to offer a high performance Mac Pro, a third party GPU (or external compute chip) is a must for the next few years. I just see them needing a couple of years to offer up an ARM based chip that is similar in performance to Fijitsu's chip. Even then an accelerator chip can still make sense no matter how fast your CPU is. So instead of dropping AMD, I rather see them increasing their partnership with AMD. In this case that might mean new technology in Mac compute accelerations. Think a fabric connection to an external CDNA chip.
This is one of the reasons Apple signed a new licensing deal with Imagination Technologies
"But as 2020 kicks off, Imagination unexpectedly announced today that Apple has signed a new multi-year licensing agreement — a deal that is deliberately being left ambiguous in the company’s press release, but will likely wind up being extremely important to upcoming Apple devices. Under the new agreement, Apple will gain “wider range” access to Imagination’s IP going forward, and since the British company’s most significant new IP focuses on a realistic graphics technology called ray tracing, it’s highly likely that Apple plans to add ray-tracing capabilities to its chips in the foreseeable future."
See the thing that worries me most are the little things in apple's actions already. If they have behind the scenes test hardware for all their configurations already, and yet they couldn't manage to get to get Thunderbolt into their developers machine. Does that mean thunderbolt is gone for them, or they couldn't get it working? Either way is not a good thing considering how much they have leaned on Thunderbolt for pro work over the past years.
And the whole idea of no external graphics card, just everything on one the chip again means less chance of expansion and more chance of just having to keep upgrading the machines every few years. The thing that was so great about the old macpro was the ability upgrade the graphics card, and graphics on the chip are not going to be upgradeable.
And then there is the whole getting companies to support it, it is great that Adobe is working on it, but what about AVID. It took them this long to support Catalina and that is with AMD graphcis.
I really don't get why people stress out about the capability of the DTK. Honestly how much clearer does Apple have to state that the machine does not represent shipping hardware???? Really it blows the mind that people can't understand the simple statements that come from Apple.
As for the no external graphics card where in the hell does it say anything about that in the posted graphic??? Now I would fully expect Apple to not stay with the status quo on the Mac Pro when it comes to expansion, The fact is expansion has done nothing to accelerate Mac Pro sales. If any thing they have a dog of a machine on their hands.
The A12Z Bionic is up to 8 GPU cores. The Most powerful and expensive GPUs have cores in the thousands. What would it take for Apple to create its own separate 500 core GPU SoC or maybe only a 100 core GPU with the ability to use several of them in a blade setup. There's nothing stopping Apple, other than patents, from making whatever they want to any way they want to. Look at the Mac Pro. It's a fantastic workstation.
That is true but cluster cored systems cost millions. The market is expanding at that end especially for AI etc. IMO Apple would be wise to cater to that market but they have zero credibility with those types of machines and a terrible record of being uncommitted to big industry projects. They would be better off creating a specific division for machines with thousands of cores (if that is what you were referencing) and letting it serve the market as a niche line.
If they come up with some new port it instead of PCI it will likely only mean Apple cards work with it, and it will mean the expansion is a joke. And while an Apple GPU might be decent, it is not going to compete with AMD and NVIDIA mainly because of driver support, because of the smaller market share unless Apple can magically beat the GPU giants at their own game without making something super hot and without a huge power draw, that is also so much easier to program for.
Oh goody this again. Happens every time.
"Apple is not going to beat out market leaders."
You clearly didn't understand what he was saying and just went on the defensive.
His comments are more than valid given current market realities and Apple's own experience and history.
We don't know where this will go. Many things are possible and nothing is sure to be a guaranteed success.
If they come up with some new port it instead of PCI it will likely only mean Apple cards work with it, and it will mean the expansion is a joke. And while an Apple GPU might be decent, it is not going to compete with AMD and NVIDIA mainly because of driver support, because of the smaller market share unless Apple can magically beat the GPU giants at their own game without making something super hot and without a huge power draw, that is also so much easier to program for.
Oh goody this again. Happens every time.
"Apple is not going to beat out market leaders."
Instead of just trolling the posters, how about comment on the substance. So you think Apple can beat high end GPU's from NVIDIA and AMD? What makes you think so? I would love for it to happen. Seems to me that supporting 3rd party graphics at least in the short term would be smart as Apple works to make some of it's own GPU hardware, but the chart does not make that seem likely.
Simple. Apple are the only company that can bring the hardware to the software i.e. they can customise the GPU ISA for Metal2. General purpose mathematics is a clumsy, power-hungry way of resolving graphics. Expect high performance from low TFLOP counts that GPU compute benchmarks will never show.
Or you could read it as that this is what the Apple Silicon processor is directly compatible with. It doesn't say they will not support third-party. I could see Apple using their own GPU for primary tasks but third-party for more beefy tasks. Do we really think that Apple could compete with a GPU like the AMD Radeon Pro Vega II for professionals?
Well this is Apple we are talking about, one whose ego knows no bounds. After all Apple did convince themselves that real pros as they are would be perfectly content with the abysmal disaster that was their butterfly keyboard(s) or how they made a bet in 2013 with the trash can Mac Pro with GPU options.
It would seem that, excluding the Mac Pro, in the future all their Macs would come equipped with integrated graphics on Apple Silicon SOC. While not a bad thing at all, as devices like the XBox and PS4 are perfect functional examples of SOCs working fine, these comehcome equipped with fast GDDR6 memory, and my question is howohow Apple handles fast bandwidth memory on their Macs
Yes, the Mac Pro tube assumed multi-GPGPU software would happen in 2013 not 2017. They have their own wide memory architecture but they could always use 1TBps HBM2 which could be great for CPU/Neural too & no round-tripping.
I'm finding it extraordinary difficult to believe that Apple could replace the existing Xeon-based Mac Pro (or even iMac Pro) with comparable performance & expandability in only two years. Low end, no problem at all.
You realise Intel haven’t had a trophy in the supercomputer cabinet for a while. Power9 has just been pipped by a 64-core ARM-based Fujitsu machine. Xeon’s done.
If they come up with some new port it instead of PCI it will likely only mean Apple cards work with it, and it will mean the expansion is a joke. And while an Apple GPU might be decent, it is not going to compete with AMD and NVIDIA mainly because of driver support, because of the smaller market share unless Apple can magically beat the GPU giants at their own game without making something super hot and without a huge power draw, that is also so much easier to program for.
Oh goody this again. Happens every time.
"Apple is not going to beat out market leaders."
You clearly didn't understand what he was saying and just went on the defensive.
His comments are more than valid given current market realities and Apple's own experience and history.
We don't know where this will go. Many things are possible and nothing is sure to be a guaranteed success.
@Beats didn’t say anything about it being a success or not. He merely pointed out that “Apple can’t compete with /*insert industry stalwart that Apple will run roughshod over in a market they couldn’t possibly think they could compete in*/” statement has been heard a number of times before.
Most famously with the iPhone, and then again with the Apple Watch.
Saying they cannot compete with AMD and NVIDIA is even more non-sensical because they’re not even going to try. Apple is not aiming to produce a graphics subsystem that has to have to world-class performance on a unlimited range of machines that they have no access to.
They are trying to produce world-class performance on a graphics system they’ve designed, running on an architecture they designed, running an operating system optimised for the chips they designed, and running software built using a framework optimised for the operating system optimised for the chipset they designed – all from the ground up. That’s why it would be unwise to count them out.
What the AMD fans will do is post loads of charts and calculations and satisfy themselves that the AMD solution is faster on paper.
Apple will just run a Windows machine and an Apple Silicon Mac side-by-side and say, “There ya go.”
See the thing that worries me most are the little things in apple's actions already. If they have behind the scenes test hardware for all their configurations already, and yet they couldn't manage to get to get Thunderbolt into their developers machine. Does that mean thunderbolt is gone for them, or they couldn't get it working? Either way is not a good thing considering how much they have leaned on Thunderbolt for pro work over the past years.
And the whole idea of no external graphics card, just everything on one the chip again means less chance of expansion and more chance of just having to keep upgrading the machines every few years. The thing that was so great about the old macpro was the ability upgrade the graphics card, and graphics on the chip are not going to be upgradeable.
And then there is the whole getting companies to support it, it is great that Adobe is working on it, but what about AVID. It took them this long to support Catalina and that is with AMD graphcis.
The A12z already worked with USB-C so it was a no brainer for the developer hardware and there was no point adding TB since I'm sure they're going to use a new chip for the production unit. I'm guessing eGPUs will work over TB with drivers once they are available but that's anybodies guess.
I'm sure Adobe and Avid will come along for the ride once they see how this new hardware outperforms x86. Technically all of their software will already run with a few days of modifications, so they'll already bee there. Xcode is processor agnostic so it's not like they have to rewrite it from the ground up and they'll be able to optimize over time as new version come out.
How are developers/manufacturers of thunderbolt audio interfaces supposed to test supporting Apple Silicon?
What I've been noticing is that there are a lot of people that read into Apple's documentation things that Apple has never said.
The document here was presented in a video where the presenter said that Intel Macs have Intel, Nvidia or AMD GPUs, Apple Silicon Macs will have Apple GPUs. I'm sure he would have said Apple Silicon Macs will have Apple GPUs and other GPUs if that was going to be the case because it's a video for developers telling them how they should design their software.
In this case we have people seemingly taking a statement that says nothing about discreet GPU's as a statement that Apple has no intention to use AMD GPU's in the future. It doesn't matter what Apple is doing in this case because it is a stretch to imply that AMD is out of the picture. In fact I just spent some time listening to a WWDC blurb on ray tracing support in Metal, this implies to me that they have plans that will likely require external hardware to achieve max performance. I say that because I don't see Apple having enough die space to be able to offer up a high performance ray tracing support. Now Apple could offer an external GPU or even a MCM but even the MCM could be supporting a third party chip. There are all sorts of options here.
There are options and there are ARM systems with PCIe support and 3rd party GPU support:
but 3rd party GPUs would only be needed if Apple's GPUs don't perform well enough and they seem to be suggesting they will. It's the same reason they are switching CPU.
They are also tuning Metal to work best on their GPU hardware, which has a different rendering structure than other GPUs. Apple is sending a clear message here, that they can beat everyone in silicon design so they have no need to buy 3rd party hardware and resell it when then they can do a better job themselves.
I still think that if apple wants to offer a high performance Mac Pro, a third party GPU (or external compute chip) is a must for the next few years. I just see them needing a couple of years to offer up an ARM based chip that is similar in performance to Fijitsu's chip. Even then an accelerator chip can still make sense no matter how fast your CPU is. So instead of dropping AMD, I rather see them increasing their partnership with AMD. In this case that might mean new technology in Mac compute accelerations. Think a fabric connection to an external CDNA chip.
The Afterburner card is a PCIe card so if they want to keep using that, they need some kind of interface for it. But with a custom GPU, they could build this into a higher-end package in the iMac Pro without needing a card or the iMac Pro can have an internal slot for it.
I suspect they will leave the Mac Pro on Intel hardware. They should be able to match its performance in the iMac form factor and over time people will wonder why pay the Intel premium of thousands extra per CPU when a standard iMac can do the same job at a much better price.
It's possible that higher-end AMD/Nvidia GPUs will outperform some of Apple's options if Apple sticks to lower power profiles but it doesn't really matter. Once you have a GPU that is over 10TFLOPs, everything beyond that is gravy. Gravy is nice but optional and it doesn't matter as much for real-time processing as for offline processing for which multiple machines can be used.
For hardware lineup, I'm expecting: - Macbook Air and mini 8-core (4-big, 4 small) CPU, 4TFLOP GPU (MBP level performance) - Macbook Pro 16-core (could be 12 big, 4 small), 8TFLOP GPU (iMac level performance) - iMac 32-core, 16TFLOP GPU (Mac Pro level performance), possibly with optional accelerators like Afterburner
- Mac Pro continue to use Intel Xeon and 3rd party GPUs, updated every 2-3 years and probably removed from sale in 6-8 years
If that's what they manage at launch, there's nothing in the roadmap that will need them to use 3rd party hardware any more but we'll only know for sure when they start shipping the first ARM Mac (likely Macbook Air) later in the year.
Comments
The current fastest Mac Pro graphics option is 56TFLOPs as it uses quad 14TFLOP GPU, which costs over $10k. Reaching that level of performance in a single chip would be an achievement but isn't necessary because the pace of mobile chip development is faster. If they can do 16TFLOPs this year and 24-32TFLOPs in a couple of years, this kind of performance is enough for any professional workflow and they can supplement it with custom accelerators like they do with the AfterBurner and T2 chips.
I'd say the software differences will cause more issues. That support doc says they won't support OpenCL on the CPU:
https://developer.apple.com/documentation/xcode/porting_your_macos_apps_to_apple_silicon
It's good they will have OpenGL and OpenCL support so all that software can be ported over but OpenCL is only supported on the GPU. In theory that's ok because OpenCL usually runs faster there but often GPUs are less reliable for compute tasks. Intel demoed a 25x speedup using OpenCL on the CPU:
https://www.khronos.org/assets/uploads/developers/library/2010_siggraph_bof_opencl/OpenCL-BOF-Intel-SIGGRAPH-Jul10.pdf
Hopefully Apple's GPU OpenCL implementation on their own hardware will be reliable but it hasn't been on AMD GPUs. If it's not reliable and developers have to fall back to unoptimized CPU code, that could result in some software running much more slowly. There's no way to know until people start testing this kind of software but it's an unusual omission leaving out OpenCL support from the CPU.
In this case we have people seemingly taking a statement that says nothing about discreet GPU's as a statement that Apple has no intention to use AMD GPU's in the future. It doesn't matter what Apple is doing in this case because it is a stretch to imply that AMD is out of the picture. In fact I just spent some time listening to a WWDC blurb on ray tracing support in Metal, this implies to me that they have plans that will likely require external hardware to achieve max performance. I say that because I don't see Apple having enough die space to be able to offer up a high performance ray tracing support. Now Apple could offer an external GPU or even a MCM but even the MCM could be supporting a third party chip. There are all sorts of options here.
I still think that if apple wants to offer a high performance Mac Pro, a third party GPU (or external compute chip) is a must for the next few years. I just see them needing a couple of years to offer up an ARM based chip that is similar in performance to Fijitsu's chip. Even then an accelerator chip can still make sense no matter how fast your CPU is. So instead of dropping AMD, I rather see them increasing their partnership with AMD. In this case that might mean new technology in Mac compute accelerations. Think a fabric connection to an external CDNA chip.
In any event you make a fundamental mistake here, Apple has been working on these chips for years now. I would imagine that they have already taped out more that a few models already.
As for the no external graphics card where in the hell does it say anything about that in the posted graphic??? Now I would fully expect Apple to not stay with the status quo on the Mac Pro when it comes to expansion, The fact is expansion has done nothing to accelerate Mac Pro sales. If any thing they have a dog of a machine on their hands.
"But as 2020 kicks off, Imagination unexpectedly announced today that Apple has signed a new multi-year licensing agreement — a deal that is deliberately being left ambiguous in the company’s press release, but will likely wind up being extremely important to upcoming Apple devices. Under the new agreement, Apple will gain “wider range” access to Imagination’s IP going forward, and since the British company’s most significant new IP focuses on a realistic graphics technology called ray tracing, it’s highly likely that Apple plans to add ray-tracing capabilities to its chips in the foreseeable future."
https://venturebeat.com/2020/01/02/apples-imagination-technologies-deal-is-all-about-ray-tracing-and-ar/
His comments are more than valid given current market realities and Apple's own experience and history.
We don't know where this will go. Many things are possible and nothing is sure to be a guaranteed success.
Expect high performance from low TFLOP counts that GPU compute benchmarks will never show.
They have their own wide memory architecture but they could always use 1TBps HBM2 which could be great for CPU/Neural too & no round-tripping.
Xeon’s done.
There are options and there are ARM systems with PCIe support and 3rd party GPU support:
https://www.anandtech.com/show/15733/ampere-emag-system-a-32core-arm64-workstation
but 3rd party GPUs would only be needed if Apple's GPUs don't perform well enough and they seem to be suggesting they will. It's the same reason they are switching CPU.
They are also tuning Metal to work best on their GPU hardware, which has a different rendering structure than other GPUs. Apple is sending a clear message here, that they can beat everyone in silicon design so they have no need to buy 3rd party hardware and resell it when then they can do a better job themselves.
The Afterburner card is a PCIe card so if they want to keep using that, they need some kind of interface for it. But with a custom GPU, they could build this into a higher-end package in the iMac Pro without needing a card or the iMac Pro can have an internal slot for it.
I suspect they will leave the Mac Pro on Intel hardware. They should be able to match its performance in the iMac form factor and over time people will wonder why pay the Intel premium of thousands extra per CPU when a standard iMac can do the same job at a much better price.
It's possible that higher-end AMD/Nvidia GPUs will outperform some of Apple's options if Apple sticks to lower power profiles but it doesn't really matter. Once you have a GPU that is over 10TFLOPs, everything beyond that is gravy. Gravy is nice but optional and it doesn't matter as much for real-time processing as for offline processing for which multiple machines can be used.
For hardware lineup, I'm expecting:
- Macbook Air and mini 8-core (4-big, 4 small) CPU, 4TFLOP GPU (MBP level performance)
- Macbook Pro 16-core (could be 12 big, 4 small), 8TFLOP GPU (iMac level performance)
- iMac 32-core, 16TFLOP GPU (Mac Pro level performance), possibly with optional accelerators like Afterburner
- Mac Pro continue to use Intel Xeon and 3rd party GPUs, updated every 2-3 years and probably removed from sale in 6-8 years
If that's what they manage at launch, there's nothing in the roadmap that will need them to use 3rd party hardware any more but we'll only know for sure when they start shipping the first ARM Mac (likely Macbook Air) later in the year.