7 ms·

Apple Executive Discusses New Mac Pro's Lack of Graphics Card Support

55 points by SenHeng 3y ago

smoldesu 3y ago
> "Fundamentally, we've built our architecture around this shared memory model and that optimization, and so it's not entirely clear to me how you'd bring in another GPU and do so in a way that is optimized for our systems," So... why expose PCI lanes in the first place, then? "Optimization" isn't really an excuse, as long as you can saturate the PCI bus it should run smoothly. Some support would be better than nothing, especially if you're brave enough to pay $7,000 for a Mac Pro. This whole thing is kinda baffling. Surely they haven't gotten rid of the multi-GPU address space support, they still support Intel/AMD systems after all.
- pavlov 3y ago
  I think quite a few people would be happy to be able to put an Nvidia card in one of those Mac Pro slots and use it through CUDA only, as a compute/ML accelerator. But Apple doesn't want to allow that either because they see CUDA as a competitor to their own APIs. So they do this whole spiel about the indivisible beauty of their own shared memory architecture.
  - pyrolistical 3y ago
    I don’t get how Apple is going to block nvidia from releasing cuda only drivers?
    - fzzzy 3y ago
      Nvidia hasn't released a Mac driver in years due to Apple's dislike for them, so they effectively already did block them.
- shrubble 3y ago
  What I don't get is, why the lowered RAM limit? Wouldn't being able to run a 512GB model on the GPU be a big selling point?
  - jitl 3y ago
    I think it’s a symptom of the same system design - Apple needs the RAM on the chiplet or as close as possible, because it’s shared with the GPU and wiring the GPU up to standard DIMMs all the way out on a motherboard would yield unacceptable latency. These are essentially phone CPUs that got super super beefy - they’re limited by what they can fit in their chiplet/system-on-chip. There’s no allowance for heterogeneous hardware access like a second tier of RAM via a motherboard northbridge or something. I think they’ll get to those higher RAM densities by M4 or M5 once they have more die space to spend on memory controllers.
  - villgax 3y ago
    For M5 maybe
sh34r 3y ago
I suspect Apple would be perfectly content if they never sold another Mac Pro. I suspect most Mac Pros manufactured are for internal use. Their main products don't have a GPU, so neither does this. It already had an F-U price. Whatever customers use them for, it's an irrelevant side hustle for Apple. I'd say they should just kill the product off to stop all the incessant whining about what a ripoff it is, but I'm sure they'd counter that there's no such thing as bad publicity.
- seanmcdirmid 3y ago
  Their main products have integrated GPUs (well, on chip), just not discrete ones. It is an important distinction, especially since they’ve been seen as a winner for portable applications.
superkuh 3y ago
Apple consistently fails to fully implement hardware protocols because they only need to implement the subset that their hardware and peripherals use. This is yet another example in a very long list for the M1/M2 series. They're cut corner hardware.
- jitl 3y ago
  They’re big phones. Why implement all that legacy stuff for a phone? I image they get a ton from simplifying their hardware engineering from supporting both the PC-style architecture that necessitates supporting a bunch of interop AND the phone style architecture that necessitates close integration to only doing things phone-style.
sacnoradhq 3y ago
Apple executive 1: "Let's add last gen PCIe slots that don't do anything." Apple executive 2: "Great! What are they for?" Apple executive 1: "They're decorations to make them look like prosumer systems with the hope of future expansion." Apple executive 2: "Okay. Then what will we promise?" Apple executive 1: "That's a tomorrow problem. We're selling hope here."
hdygdtfd 3y ago
history repeats itself. it's 1989 all over again.
bastard_op 3y ago
You'll take what Apple gives you, or go back your adventures in plebian life. Who needs an industry-leading gpu for graphics or compute. I'm curious to see how far gaming goes with their translation+gpu vs. likes of nvidia. Same problems linux has with wine, translation against windoze directx, or lack thereof.
- lockhouse 3y ago
  People that want this from a computer won’t buy a Mac. It’s like complaining that a tractor is too slow to drive down the interstate.
AraceliHarker 3y ago
If expandability is not what Apple wants, then what is the space inside Mac Pro for?
- readthenotes1 3y ago
  Allows for lower energy consumption per cubic foot. Planet saving dontcha know?
- Tagbert 3y ago
  Non-GPU PCI cards. A lot are I/O for audio and other scenarios.
photonbeam 3y ago
What actually stops amd/nvidia form slotting a card into it, writing a driver, and letting people use it?
- JohnTHaller 3y ago
  Same thing that stopped Nvidia from releasing a driver for their modern cards for Intel Macs... Apple.
- fuzzythinker 3y ago
  their bottom line
imagetic 3y ago
The Mac Pro has a lot of significance in the professional media industry. Final Cut changed the landscape forever and at $3500 entry level Mac Pro that could compete with a $70k AVID station after a $1000 software purchase and an Nvidia card upgrade. Now as a college student I could go work for myself because I could (barely) afford the tools. I still work on a Mac to this day. But that $3500/entry level and upgrade into your career machine isn't the goal in whatever the heck these are. Aside from maybe like a Hollywood editor who wants to use a mac and gets their way, I have no idea who these are for. I think the goal is to just look really cool in a fancy office.
- goerz 3y ago
  > Aside from maybe like a Hollywood editor […], I have no idea who these are for. That’s exactly who these are for
- Tanoc 3y ago
  I've noticed dedicated Mac rendering stations have been getting rarer and rarer in favour of Windows or Linux builds. Small businesses are buying multi-core CPU systems and then upgrading the RAM and GPU every four or five years now instead of buying an entirely new Mac Pro every four to five years. A print shop I went to recently had a Windows machine running 32GB of RAM and a 3060, whereas previously that would've been entirely the domain of a Mac.
musictubes 3y ago
When pressed by Gruber, Ternus says that there are PCI cards made specifically for Avid, Pro Tools, video capture cards, networking, and even some custom hardware configurations. The Mac Pro can also be loaded up with a lot of PCIe flash storage. So yeah, a niche machine but there is a lot more to PCI than graphics cards.
- sudosysgen 3y ago
  You can most of that via USB, though. And for storage you'd be better served by M.2.
  - musictubes 3y ago
    Professional AV rigs can get quite hairy. This video shows how one composer’s computer setup went from a pile of boxes and cables to a single rack mounted machine. I’ve set the playback point to where he shows the difference. https://youtu.be/xNrG2mwt4Uo?t=1995 https://youtu.be/xNrG2mwt4Uo?t=1995 And in this 4 minute video you can see all of his cards installed into his Mac Pro with nary a GPU in sight. https://youtu.be/kIQINCWMd6I https://youtu.be/kIQINCWMd6I And he is just a single person. Imagine a company with 7-10 or more of those machines for both audio and video production. Desktop “workstation” computers have become relics of a bygone age. Having one or two external graphics cards is really only relevant to gamers and hobbyist ML tinkerers. Actual model training is done on the cloud. The few people/organizations that actually use PCI cards in a desktop will be well served by the new Mac Pro.
  - DrBenCarson 3y ago
    You would but there’s a finite amount of M.2 you can put on a motherboard. NVME is just PCIE at the end of the day so expanding storage is a real use for these slots.
musictubes 3y ago
https://www.avid.com/products/pro-tools-hdx https://www.avid.com/products/pro-tools-hdx https://www.newegg.com/p/pl?d=pci+fiber+network+card https://www.newegg.com/p/pl?d=pci+fiber+network+card https://www.blackmagicdesign.com/products/decklink/techspecs/W-DLK-34 https://www.blackmagicdesign.com/products/decklink/techspecs... https://www.pcworld.com/article/394512/the-best-pcie-40-ssd.html https://www.pcworld.com/article/394512/the-best-pcie-40-ssd.... So yeah, high end audio and video workflows can use plenty of PCI cards. They’re more to PCI than GPUs.
__d 3y ago
Early personal computers used a basically passive backplane (eg. S-100 bus). After that, the Apple ][, IBM PC, and later Macintosh models used a motherboard with expansion cards. Throughout this era (from, say 1974 through until 2022), the elements that were composed to create a personal computer were MSI, LSI, and VLSI integrated circuits mounted onto a PCB, and wired up with traces on the board(s) and expansion card slots. The M1-based Macs introduced a new era in personal computer architecture: previously used for phones and embedded devices, the SoC, and particularly, the chiplet-based SoC, took over from the motherboard. The elements composed to make a PC now are chiplets, wired up with silicon interposers, and encapsulated into an SMT chip. The benefits in speed, bandwidth, and power usage of a SoC over VLSI-on-a-PCB are enormous, and reflected in the performance of the M1-based Macs. Where do expansion cards fit with an SoC-based model? They're slow, narrow bandwidth, and power-hungry devices. A GPU expansion card on an SoC-based computer might as well be accessed over an Ethernet. Of course, it's disruptive. People legitimately _like_ the ability to compose their own special combination of functions in their PC, and it allows a level of _end-user_ customization that isn't (currently?) possible with SoCs. But retaining that level of customization means significant performance costs, and the gap is only going to grow as further "3D" chiplet assembly techniques become common. The cost of creating a SoC from chiplets is sufficiently high that a market of one (ie. end-user customization) isn't possible. Right now, we get base, Pro, Max, and Ultra variants from Apple. It's possible we'll get more in future, but ... it's fundamentally a mass-market technology. The era of end-user hardware customization is very likely drawing to a close.
- zamadatix 3y ago
  That the SoC has the same PCIe 4.0 x16 bus as a typical PC has doesn't seem to be an inherent issue in utilizing more powerful GPUs itself. After all, we stick more powerful GPUs on typical PCs and they outclass the M2 Ultra in many workloads just fine, slow bus be damned. Being closely integrated with the rest of the SoC definitely has its benefits but it often isn't all that important unless your workload specifically needs to share across CPU+RAM+GPU+VRAM nearly constantly. Which for laptop stuff or media editing, sure - it fits that use case perfectly and I see why they'd never want to start copying the raw video files over a bus to the GPU just to run it through the encoder and copy it back to main memory. I think the more interesting question is: what additional would having add-on GPU support realistically provide for this type of product? Enable you to play a limited subset of games on a ~$10,000 Mac Workstation which will run them no better than a standard PC which can play anything? Enable you train AI models faster or kerchunk GPGPU faster while connected to really expensive SoC with no other purpose to contain more peripheral I/O? All while you use an OS that's just extra inconvenient for the task vs. the Linux environments everyone is already using? A Mac workstation is already a niche and if they have something that can target the main use case of that niche (media production) perfectly already then putting as much work into designing new products for additional niches of the already most niche use case just isn't worth it. Particularly when the only reasonably priced way would be to get rid of the integrated solution that serves the main use case better. Not that the technology couldn't be made to do it as good as any other system could, it just doesn't make sense to go after since it's so far from their markets. Really the M series is already an expansion of "well our phone chip is fast..." as is. It has a hard enough time adapting to being a useful laptop (e.g. display outputs) let alone replace every type of system.
gentleman11 3y ago
An apple event; they bring Kojima in to talk about gaming on Mac; they introduce gaming mode; they are excited by gaming on Mac; they still don't offer dedicated gpus? Not even in their workstations to try to appeal to game developers? I don't know what to say. Is it because gpus are upgradeable, and they are allergic to this, retreating from it like vampires from holy water? Is it to keep the price of GPUs from muddying the water in their careful pricing ladder where the cpu and gpu are bundled? It's insulting
- abhibeckert 3y ago
  I'm not much into gaming on a Mac — most of the games I like to play only run on Windows, and I have a gaming PC for that... But I have tried a few games on the Mac laptop I use for work, which doesn't do anything graphics intensive so it has the slowest Apple GPU available (as in, the first one where they stopped using Intel or AMD GPUs). Fore very game I've tested (not many admittedly) I can turn the graphics settings up to 11 and it they run without dropping frames at the native resolution and refresh rate of the display (3K @ 60Hz) even on battery power in "low power mode" which substantially under-clocks the GPU and CPU to improve battery life. Obviously high end GPUs like the NVIDIA A100 are significantly faster, but I'm pretty sure everyone using those on a gaming PC is doing it for bragging rights and nothing else - and those people want to build their own gaming rig. Pretty much nobody would upgrade a modern Mac if it was possible. The built in GPUs are perfectly fine even on the lowly M1, let alone an M2 Ultra. It's just not worth the engineering effort (by Apple or by GPU manufacturers) for the small number of customers who'd be willing to spend the money. And honestly even the A100 would be a downgrade in some ways. NVIDIA's flagship has an order of magnitude less memory than the M2 Ultra for example.
  - kkielhofner 3y ago
    Maybe you’re confusing an A100 for something like an RTX 4090? No one is using an A100 for gaming. They’re > $10k. They don’t even have video out. They’re also actually slower for most gaming and related tasks - the RTX 4090 is juiced to the gills power and heat be damned all to push frames as fast as possible. The A100 is heavily optimized for dense datacenter applications where power and heat are among the top concerns. They also have up to 80GB of RAM, and the RTX 4090 has 24GB which is a ways away from “an order of magnitude” less memory when the M2 Ultra supports a max of 192GB of unified memory.
hyperhello 3y ago
Maybe Apple will release PCI expansion cards itself that you can take from Pro to Pro.
villgax 3y ago
There is no excuse to not support eGPUs though
bentt 3y ago
This is the long term result of Apple not wanting any further dependency on AMD/Nvidia, but at the same time not realizing that AI was going to hit and be very, very Nvidia centric. But for most tasks which were known to exist 5+ years ago when these decisions were made, their on die gpu really is plenty fast and offers a larger memory area than most every gpu known to be coming at that time.
- foverzar 3y ago
  Apple has AI hardware acceleration in their devices. Nvidia's advantages in AI come from industrial grade hardware that lives in datacenters, and Apple hasn't been competitive in datacenter hardware for decades.
SilverBirch 3y ago
A lot of people in these comments don't seem to be really engaging with what was said. The unified memory architecture is fantastic, everything is in one place, easy and fast to access, up to the limit of the size of memory and compute you can fit on one chip it's massively advantageous. There is obviously a point where you reach the limit of how much you can pack in, but the disadvantages of having to shovel data back and forth mean you need a really powerful GPU before it becomes worth having. But let's assume you really want more compute Apple has to first design a standalone GPU, then it has to do a PCI or CXL core or some custom interconnect, then it has to redesign it's OS for non-uniform memory doing all that fun stuff hiding the memory communication, then it has to add the primitives to it's languages for you to manage memory (because to get performance you have to actively design for it) and then you need to get all your performance critical applications to re-write their stacks with this in mind. It's a massive lift! And for what? Only Desktop can take advantage of it, and in Apple's product line only the Mac Pro. And that's assume you execute well and the device ends up more capable than just using the SoC. And all of this we're talking about on top of the massive lift Apple literally just finished moving all their products onto Apple silicon. It's just a crazy ask for a v1 of a product.
- bogantech 3y ago
  > then it has to redesign it's OS for non-uniform memory doing all that fun stuff hiding the memory communication, then it has to add the primitives to it's languages for you to manage memory (because to get performance you have to actively design for it) and then you need to get all your performance critical applications to re-write their stacks with this in mind. Yeah, it's not like their OS and Languages already support this model given that everything before the switch to Arm worked this way
  - Hardwired8976 3y ago
    Before it was just like on windows with a graphics card that holds the compute relevant data. Now they have two cards with different memory bandwidths and ideally you would want to combine them of some sorts instead of wasting the whole m2 ultra gpu
- 3y ago
andrewmcwatters 3y ago
I suspect there is a more fundamental truth here that’s practically reliable. Apple wants you to upgrade your hardware by purchasing new devices, not components, because that sells more units. If you think about all their insulting decisions from that lens, everything else makes more sense than any other basic assumption.

55 points by SenHeng 3y ago

> "Fundamentally, we've built our architecture around this shared memory model and that optimization, and so it's not entirely clear to me how you'd bring in another GPU and do so in a way that is optimized for our systems," So... why expose PCI lanes in the first place, then? "Optimization" isn't really an excuse, as long as you can saturate the PCI bus it should run smoothly. Some support would be better than nothing, especially if you're brave enough to pay $7,000 for a Mac Pro. This whole thing is kinda baffling. Surely they haven't gotten rid of the multi-GPU address space support, they still support Intel/AMD systems after all.

pavlov 3y ago
I think quite a few people would be happy to be able to put an Nvidia card in one of those Mac Pro slots and use it through CUDA only, as a compute/ML accelerator. But Apple doesn't want to allow that either because they see CUDA as a competitor to their own APIs. So they do this whole spiel about the indivisible beauty of their own shared memory architecture.
- pyrolistical 3y ago
  I don’t get how Apple is going to block nvidia from releasing cuda only drivers?
  - fzzzy 3y ago
    Nvidia hasn't released a Mac driver in years due to Apple's dislike for them, so they effectively already did block them.
shrubble 3y ago
What I don't get is, why the lowered RAM limit? Wouldn't being able to run a 512GB model on the GPU be a big selling point?
- jitl 3y ago
  I think it’s a symptom of the same system design - Apple needs the RAM on the chiplet or as close as possible, because it’s shared with the GPU and wiring the GPU up to standard DIMMs all the way out on a motherboard would yield unacceptable latency. These are essentially phone CPUs that got super super beefy - they’re limited by what they can fit in their chiplet/system-on-chip. There’s no allowance for heterogeneous hardware access like a second tier of RAM via a motherboard northbridge or something. I think they’ll get to those higher RAM densities by M4 or M5 once they have more die space to spend on memory controllers.
- villgax 3y ago
  For M5 maybe

sh34r 3y ago

I suspect Apple would be perfectly content if they never sold another Mac Pro. I suspect most Mac Pros manufactured are for internal use. Their main products don't have a GPU, so neither does this. It already had an F-U price. Whatever customers use them for, it's an irrelevant side hustle for Apple. I'd say they should just kill the product off to stop all the incessant whining about what a ripoff it is, but I'm sure they'd counter that there's no such thing as bad publicity.

seanmcdirmid 3y ago
Their main products have integrated GPUs (well, on chip), just not discrete ones. It is an important distinction, especially since they’ve been seen as a winner for portable applications.

superkuh 3y ago

Apple consistently fails to fully implement hardware protocols because they only need to implement the subset that their hardware and peripherals use. This is yet another example in a very long list for the M1/M2 series. They're cut corner hardware.

jitl 3y ago
They’re big phones. Why implement all that legacy stuff for a phone? I image they get a ton from simplifying their hardware engineering from supporting both the PC-style architecture that necessitates supporting a bunch of interop AND the phone style architecture that necessitates close integration to only doing things phone-style.

sacnoradhq 3y ago

Apple executive 1: "Let's add last gen PCIe slots that don't do anything." Apple executive 2: "Great! What are they for?" Apple executive 1: "They're decorations to make them look like prosumer systems with the hope of future expansion." Apple executive 2: "Okay. Then what will we promise?" Apple executive 1: "That's a tomorrow problem. We're selling hope here."

hdygdtfd 3y ago

history repeats itself. it's 1989 all over again.

bastard_op 3y ago

You'll take what Apple gives you, or go back your adventures in plebian life. Who needs an industry-leading gpu for graphics or compute. I'm curious to see how far gaming goes with their translation+gpu vs. likes of nvidia. Same problems linux has with wine, translation against windoze directx, or lack thereof.

lockhouse 3y ago
People that want this from a computer won’t buy a Mac. It’s like complaining that a tractor is too slow to drive down the interstate.

AraceliHarker 3y ago

If expandability is not what Apple wants, then what is the space inside Mac Pro for?

readthenotes1 3y ago
Allows for lower energy consumption per cubic foot. Planet saving dontcha know?
Tagbert 3y ago
Non-GPU PCI cards. A lot are I/O for audio and other scenarios.

photonbeam 3y ago

What actually stops amd/nvidia form slotting a card into it, writing a driver, and letting people use it?

JohnTHaller 3y ago
Same thing that stopped Nvidia from releasing a driver for their modern cards for Intel Macs... Apple.
fuzzythinker 3y ago
their bottom line

imagetic 3y ago

The Mac Pro has a lot of significance in the professional media industry. Final Cut changed the landscape forever and at $3500 entry level Mac Pro that could compete with a $70k AVID station after a $1000 software purchase and an Nvidia card upgrade. Now as a college student I could go work for myself because I could (barely) afford the tools. I still work on a Mac to this day. But that $3500/entry level and upgrade into your career machine isn't the goal in whatever the heck these are. Aside from maybe like a Hollywood editor who wants to use a mac and gets their way, I have no idea who these are for. I think the goal is to just look really cool in a fancy office.

goerz 3y ago
> Aside from maybe like a Hollywood editor […], I have no idea who these are for. That’s exactly who these are for
Tanoc 3y ago
I've noticed dedicated Mac rendering stations have been getting rarer and rarer in favour of Windows or Linux builds. Small businesses are buying multi-core CPU systems and then upgrading the RAM and GPU every four or five years now instead of buying an entirely new Mac Pro every four to five years. A print shop I went to recently had a Windows machine running 32GB of RAM and a 3060, whereas previously that would've been entirely the domain of a Mac.

musictubes 3y ago

When pressed by Gruber, Ternus says that there are PCI cards made specifically for Avid, Pro Tools, video capture cards, networking, and even some custom hardware configurations. The Mac Pro can also be loaded up with a lot of PCIe flash storage. So yeah, a niche machine but there is a lot more to PCI than graphics cards.

sudosysgen 3y ago
You can most of that via USB, though. And for storage you'd be better served by M.2.
- musictubes 3y ago
  Professional AV rigs can get quite hairy. This video shows how one composer’s computer setup went from a pile of boxes and cables to a single rack mounted machine. I’ve set the playback point to where he shows the difference. https://youtu.be/xNrG2mwt4Uo?t=1995 https://youtu.be/xNrG2mwt4Uo?t=1995 And in this 4 minute video you can see all of his cards installed into his Mac Pro with nary a GPU in sight. https://youtu.be/kIQINCWMd6I https://youtu.be/kIQINCWMd6I And he is just a single person. Imagine a company with 7-10 or more of those machines for both audio and video production. Desktop “workstation” computers have become relics of a bygone age. Having one or two external graphics cards is really only relevant to gamers and hobbyist ML tinkerers. Actual model training is done on the cloud. The few people/organizations that actually use PCI cards in a desktop will be well served by the new Mac Pro.
- DrBenCarson 3y ago
  You would but there’s a finite amount of M.2 you can put on a motherboard. NVME is just PCIE at the end of the day so expanding storage is a real use for these slots.

musictubes 3y ago

https://www.avid.com/products/pro-tools-hdx https://www.avid.com/products/pro-tools-hdx https://www.newegg.com/p/pl?d=pci+fiber+network+card https://www.newegg.com/p/pl?d=pci+fiber+network+card https://www.blackmagicdesign.com/products/decklink/techspecs/W-DLK-34 https://www.blackmagicdesign.com/products/decklink/techspecs... https://www.pcworld.com/article/394512/the-best-pcie-40-ssd.html https://www.pcworld.com/article/394512/the-best-pcie-40-ssd.... So yeah, high end audio and video workflows can use plenty of PCI cards. They’re more to PCI than GPUs.

__d 3y ago

Early personal computers used a basically passive backplane (eg. S-100 bus). After that, the Apple ][, IBM PC, and later Macintosh models used a motherboard with expansion cards. Throughout this era (from, say 1974 through until 2022), the elements that were composed to create a personal computer were MSI, LSI, and VLSI integrated circuits mounted onto a PCB, and wired up with traces on the board(s) and expansion card slots. The M1-based Macs introduced a new era in personal computer architecture: previously used for phones and embedded devices, the SoC, and particularly, the chiplet-based SoC, took over from the motherboard. The elements composed to make a PC now are chiplets, wired up with silicon interposers, and encapsulated into an SMT chip. The benefits in speed, bandwidth, and power usage of a SoC over VLSI-on-a-PCB are enormous, and reflected in the performance of the M1-based Macs. Where do expansion cards fit with an SoC-based model? They're slow, narrow bandwidth, and power-hungry devices. A GPU expansion card on an SoC-based computer might as well be accessed over an Ethernet. Of course, it's disruptive. People legitimately _like_ the ability to compose their own special combination of functions in their PC, and it allows a level of _end-user_ customization that isn't (currently?) possible with SoCs. But retaining that level of customization means significant performance costs, and the gap is only going to grow as further "3D" chiplet assembly techniques become common. The cost of creating a SoC from chiplets is sufficiently high that a market of one (ie. end-user customization) isn't possible. Right now, we get base, Pro, Max, and Ultra variants from Apple. It's possible we'll get more in future, but ... it's fundamentally a mass-market technology. The era of end-user hardware customization is very likely drawing to a close.

zamadatix 3y ago
That the SoC has the same PCIe 4.0 x16 bus as a typical PC has doesn't seem to be an inherent issue in utilizing more powerful GPUs itself. After all, we stick more powerful GPUs on typical PCs and they outclass the M2 Ultra in many workloads just fine, slow bus be damned. Being closely integrated with the rest of the SoC definitely has its benefits but it often isn't all that important unless your workload specifically needs to share across CPU+RAM+GPU+VRAM nearly constantly. Which for laptop stuff or media editing, sure - it fits that use case perfectly and I see why they'd never want to start copying the raw video files over a bus to the GPU just to run it through the encoder and copy it back to main memory. I think the more interesting question is: what additional would having add-on GPU support realistically provide for this type of product? Enable you to play a limited subset of games on a ~$10,000 Mac Workstation which will run them no better than a standard PC which can play anything? Enable you train AI models faster or kerchunk GPGPU faster while connected to really expensive SoC with no other purpose to contain more peripheral I/O? All while you use an OS that's just extra inconvenient for the task vs. the Linux environments everyone is already using? A Mac workstation is already a niche and if they have something that can target the main use case of that niche (media production) perfectly already then putting as much work into designing new products for additional niches of the already most niche use case just isn't worth it. Particularly when the only reasonably priced way would be to get rid of the integrated solution that serves the main use case better. Not that the technology couldn't be made to do it as good as any other system could, it just doesn't make sense to go after since it's so far from their markets. Really the M series is already an expansion of "well our phone chip is fast..." as is. It has a hard enough time adapting to being a useful laptop (e.g. display outputs) let alone replace every type of system.

gentleman11 3y ago

An apple event; they bring Kojima in to talk about gaming on Mac; they introduce gaming mode; they are excited by gaming on Mac; they still don't offer dedicated gpus? Not even in their workstations to try to appeal to game developers? I don't know what to say. Is it because gpus are upgradeable, and they are allergic to this, retreating from it like vampires from holy water? Is it to keep the price of GPUs from muddying the water in their careful pricing ladder where the cpu and gpu are bundled? It's insulting

abhibeckert 3y ago
I'm not much into gaming on a Mac — most of the games I like to play only run on Windows, and I have a gaming PC for that... But I have tried a few games on the Mac laptop I use for work, which doesn't do anything graphics intensive so it has the slowest Apple GPU available (as in, the first one where they stopped using Intel or AMD GPUs). Fore very game I've tested (not many admittedly) I can turn the graphics settings up to 11 and it they run without dropping frames at the native resolution and refresh rate of the display (3K @ 60Hz) even on battery power in "low power mode" which substantially under-clocks the GPU and CPU to improve battery life. Obviously high end GPUs like the NVIDIA A100 are significantly faster, but I'm pretty sure everyone using those on a gaming PC is doing it for bragging rights and nothing else - and those people want to build their own gaming rig. Pretty much nobody would upgrade a modern Mac if it was possible. The built in GPUs are perfectly fine even on the lowly M1, let alone an M2 Ultra. It's just not worth the engineering effort (by Apple or by GPU manufacturers) for the small number of customers who'd be willing to spend the money. And honestly even the A100 would be a downgrade in some ways. NVIDIA's flagship has an order of magnitude less memory than the M2 Ultra for example.
- kkielhofner 3y ago
  Maybe you’re confusing an A100 for something like an RTX 4090? No one is using an A100 for gaming. They’re > $10k. They don’t even have video out. They’re also actually slower for most gaming and related tasks - the RTX 4090 is juiced to the gills power and heat be damned all to push frames as fast as possible. The A100 is heavily optimized for dense datacenter applications where power and heat are among the top concerns. They also have up to 80GB of RAM, and the RTX 4090 has 24GB which is a ways away from “an order of magnitude” less memory when the M2 Ultra supports a max of 192GB of unified memory.

hyperhello 3y ago

Maybe Apple will release PCI expansion cards itself that you can take from Pro to Pro.

villgax 3y ago

There is no excuse to not support eGPUs though

bentt 3y ago

This is the long term result of Apple not wanting any further dependency on AMD/Nvidia, but at the same time not realizing that AI was going to hit and be very, very Nvidia centric. But for most tasks which were known to exist 5+ years ago when these decisions were made, their on die gpu really is plenty fast and offers a larger memory area than most every gpu known to be coming at that time.

foverzar 3y ago
Apple has AI hardware acceleration in their devices. Nvidia's advantages in AI come from industrial grade hardware that lives in datacenters, and Apple hasn't been competitive in datacenter hardware for decades.

SilverBirch 3y ago

A lot of people in these comments don't seem to be really engaging with what was said. The unified memory architecture is fantastic, everything is in one place, easy and fast to access, up to the limit of the size of memory and compute you can fit on one chip it's massively advantageous. There is obviously a point where you reach the limit of how much you can pack in, but the disadvantages of having to shovel data back and forth mean you need a really powerful GPU before it becomes worth having. But let's assume you really want more compute Apple has to first design a standalone GPU, then it has to do a PCI or CXL core or some custom interconnect, then it has to redesign it's OS for non-uniform memory doing all that fun stuff hiding the memory communication, then it has to add the primitives to it's languages for you to manage memory (because to get performance you have to actively design for it) and then you need to get all your performance critical applications to re-write their stacks with this in mind. It's a massive lift! And for what? Only Desktop can take advantage of it, and in Apple's product line only the Mac Pro. And that's assume you execute well and the device ends up more capable than just using the SoC. And all of this we're talking about on top of the massive lift Apple literally just finished moving all their products onto Apple silicon. It's just a crazy ask for a v1 of a product.

bogantech 3y ago
> then it has to redesign it's OS for non-uniform memory doing all that fun stuff hiding the memory communication, then it has to add the primitives to it's languages for you to manage memory (because to get performance you have to actively design for it) and then you need to get all your performance critical applications to re-write their stacks with this in mind. Yeah, it's not like their OS and Languages already support this model given that everything before the switch to Arm worked this way
- Hardwired8976 3y ago
  Before it was just like on windows with a graphics card that holds the compute relevant data. Now they have two cards with different memory bandwidths and ideally you would want to combine them of some sorts instead of wasting the whole m2 ultra gpu
3y ago

andrewmcwatters 3y ago

I suspect there is a more fundamental truth here that’s practically reliable. Apple wants you to upgrade your hardware by purchasing new devices, not components, because that sells more units. If you think about all their insulting decisions from that lens, everything else makes more sense than any other basic assumption.