Apple Executive Discusses New Mac Pro's Lack of Graphics Card Support
macrumors.com
macrumors.com
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
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.
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.
in terms of real usage, a sizeable population of users won’t actually benefit from upgradable graphics. for gaming or otherwise. even a lone A100. (how many users not only use an A100 but also installed it? yes, that’s silly.)
knowing an upgrade is possible, knowing an upgrade is sensible, knowing how to upgrade, and actually performing the upgrade (and successfully) are all steps down the upgrade path. would suspect the conversion rate on that is quite low considering the size of the funnel at the start. especially with the last few years of logistics.
as for these unified macs, if the system starts to show weakness, then yes, time to consider that it isn’t the right tool for the job, and maybe it is time to push that workload to a dedicated gaming pc (for now) or the cloud (for scale.)
and, yes, there is still option 3. maybe time for an upgrade… of the entire system. even with the benefit of a curated environment, that is a hefty cost. everything gets replaced or nothing gets upgraded. which, full circle, only matters when an upgrade matters. will it?
The M2 Ultra costs $4000+. For that money a 4090 or 7900 XTX is very much in the cards and will offer considerably better GPU performance.
You tested retro and mobile games or cant tell 30 from 60 fps. Just one example using almost 10 year old game medium details 30 fps https://www.youtube.com/watch?v=WGTCSKMz-ec. Even older Tomb raider with lower quality and resolution down to 15 fps https://www.youtube.com/watch?v=ZMPri2zgfak. M1 Max with 4K high settings 30 fps https://www.youtube.com/watch?v=wZVOo9WTIsU
apple's whole business model is completely reliant on them occupying a segment of the market that no one else can.
Remember every (GPU) configuration has to be supported to not hurt the user experience, kind of like making webpages that support 1x1 pixel screens (is there any?) to 8k.
Every configuration setting increases the test area exponentially. This is what makes Linux great (for some)!
They can test and bless a handful. Unless I'm missing something, it's not an all or nothing decision
The "don't let people shoot themselves in the foot" mindset has worked great for them.
Casual game revenue (mostly mobile) >> console game revenue >> revenue from AAA games on gamer rigs and those rigs themselves. Perhaps those greater thans should all be ">>>>". If you look at the entire (huge) gaming ecosystem, Apple is already a major player. If you look the segment I can tell you care about, it's tiny.
IIRC gaming got Nvidia off the ground but for a long time hasn't been their revenue pump.
Apple's gaming boost can really only get them one thing: if you're going to buy a PC anyway, perhaps you'll pay a little more and not buy a console. Maybe that already happens a bit in the PC world, I don't know, but for Apple anyway it makes sense since unlike almost all PCs, Apple have a differentiated product. And if they talked about AAA gaming (if they did -- I didn't bother to listen) I imagine it's like Honda fielding an F-1 team: staking out a presence in a high end, that makes the intermediate product look like a better deal. Like the Mac Pro: no faster than a product they sell for less than half the price, with a couple of extra features very few need. It drives real sales elsewhere.
Finally: Apple has a "problem" that a Bungie or Alienware (new Dell) don't: Apple's revenue is so huge that they simply can't enter into certain markets unless it's for marketing reasons (like the Mac Pro). Look at the ipad: it basically is the tablet market, and if it were a standalone product it would be one of the great business success stories of all time. But because it's in Apple's product mix it's ignored by financial analysis which is just one step above "failed product": "at least it doesn't lose money."
BTW I suspect NVidia had some gamers in its early team and has some in its culture. Apple carea a lot about Music because Steve Jobs cared a lot about music, but hasn't had gaming DNA for decades, really since the Apple II IMHO.
https://www.newegg.com/p/pl?d=pci+fiber+network+card
https://www.blackmagicdesign.com/products/decklink/techspecs...
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.
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.
Yeah, it's not like their OS and Languages already support this model given that everything before the switch to Arm worked this way
Why would they ever need that? They just need to provide the usual PCIe interface which the existing cards support.
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.
That’s exactly who these are for
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.
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.
My point is: Apple doesn't want to support external GPUs, because they've moved to a new architecture with the components of the PC integrated on the SoC, not across an off-chip PCIe bus. They've done that because the on-chip GPU has performance advantages on multiple dimensions.
I've no doubt that Apple would love to retain those customers who want 1.5TB of RAM, and 6 GPUs churning away at whatever task, but ... it's not worth building a SoC to support that niche, and it's not worth the performance compromise that an off-chip solution would imply.
The solution they've implemented in the Mac Pro, using PCIe switches to multiplex a bunch of slots across a handful of lanes, is treating PCIe as a secondary, low-performance I/O bus for things that don't need low-latency/high-bandwidth access to core CPU, GPU, and RAM. Which is fine for some stuff, but not really what you need for a GPU.
I'm personally no fan of losing the expansion card model, but the tradeoff is worth it for most of their customers.
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.
So yeah, a niche machine but there is a lot more to PCI than graphics cards.
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.
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.
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.
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.
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.
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."
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.
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.
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.
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.
It’s like complaining that a tractor is too slow to drive down the interstate.
it's 1989 all over again.