An Apple silicon GPGPU is what is needed here; display output from "iGPU" & compute/rendering output from ASi GPGPU(s)...
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
MP 7,1 Mac Pro 8.1 - WHAT IF...
- Thread starter maikerukun
- Start date
-
- Tags
- 7.1 mac pro next generation
- Sort by reaction score
An Apple silicon GPGPU is what is needed here; display output from "iGPU" & compute/rendering output from ASi GPGPU(s)...
We disagree, although you may well be right. Still, I bet you one worthless virtual dollar it means real 3rd party graphics cards (as well as some first party ones too)—likely AMD 7xxx series. The reason I think that, is likely dumb, but keeping the same case makes me feel there is pressure to keep that ability. I concede, this could easily be wishful thinking on my part.
That would essentially mean a whole new chip, just for the Mac Pro. The MBP and iMac lack the cooling for anything bigger than a Max, or perhaps Ultra. As does the Studio (if it doesn't get the chop).
Whilst it would of course still be technically possible, it would be very expensive; the business case for it therefore seems unlikely. But who knows? Maybe the M-series will eventually settle into a pattern of CPU + GPU chiplets, and the early models were just baby steps that didn't stray too far from the A-series.
Last edited:
I would dearly love to be wrong. In any case, all I'm doing is extrapolating from what's in the public domain. It could all be upended by the reveal of something that's behind closed doors at the moment.
IMO all Apple would need do is chop off the integrated parts and move them "outside" of the chip. The new chip could be built in a similar way that AMD does with Ryzen.
On the flip side, separating the "laptop" SoCs from desktop (workstation) SoCs would save money on the laptop SoC side of things...
I am targeting the Mn Max SoCs here; right now every M1 Max SoC has the UltraFabric connector, whether it is needed or not...
Spoiler, it is not needed unless two M1 Max SoCs are paired together to make a M1 Ultra SoC...!
So, price reduction by stripping the Mn Max laptop SoCs of the UltraFusion connector...
Back to the new Mn Max desktop (workstation) SoC; this could be in both the Mac Studio and the Mac Pro, in Max, Ultra, and Extreme variants; so four SKUs that can use it...
- Mn Max Mac Studio
- Mn Ultra Mac Studio
- Mn Ultra Mac Pro
- Mn Extreme Mac Pro
Another use for this "GPU-specific" SoC is an Apple silicon GPGPU card...
A way to spread out the cost of these ASi GPGPUs is by also making them work in eGPU enclosures, for use by all of the ASi Macs...
Seems like a possibility is emerging...
The UltraFusion connection would still be needed to link 2x Max's though; in fact they'd need a second one, to connect to the GPU silicon as well, unless I've misunderstood?
If the only difference between the laptop Max and the desktop Max is that the former lacks the UF bus, it hardly seems worth it. Likely better to just get the economies of scale from a single design, and just chop off / ignore UF for laptop use.
Or do you mean the laptop version has 0 UF connectors, and the desktop one has 2? In that case, perhaps the difference is significant enough to make the laptop version cheaper to make.
Either way, something like this is needed to get a better balance of CPU to GPU as you scale multiple chips. Otherwise, you wind up with an absurd amount of CPU cores, just to get a decent desktop-sized GPU.
The UltraFusion connection would still be needed to link 2x Max's though; in fact they'd need a second one, to connect to the GPU silicon as well, unless I've misunderstood?
If the only difference between the laptop Max and the desktop Max is that the former lacks the UF bus, it hardly seems worth it. Likely better to just get the economies of scale from a single design, and just chop off / ignore UF for laptop use.
Or do you mean the laptop version has 0 UF connectors, and the desktop one has 2? In that case, perhaps the difference is significant enough to make the laptop version cheaper to make.
Either way, something like this is needed to get a better balance of CPU to GPU as you scale multiple chips. Otherwise, you wind up with an absurd amount of CPU cores, just to get a decent desktop-sized GPU.
Last edited:
I'm pretty sure there is an abundance of potential material just by interviewing this Pro Group—easily a 6 episode Netflix mini series—about how years' worth of work could lead to a pro workstation with no memory card slots.
Yes, there are many types of cards. Yes, the keep changing the formats on a quarterly basis.
So just design a drop-in bay/cartridge that fits snugly and integrates a module that the customer chooses: single format? SD+CFast? CFast+CFExpress? RED Mags? 4-in-one? 8-in-one? 12-in-one?
FFS, already.
Two different Mn Max SoCs, one "laptop" and one "desktop"...
Mn Max "laptop" SoC does not need the UltraFusion connector, the Mn Max "desktop" SoC does...
"GPU-specific" SoC would be for desktops and would also have an UltraFusion connector...
The need for a second UltraFusion connector on any single die would be for a Mn Extreme SoC, which is four Mn Max SoCs "strapped together"...
Variant would be the "regular" desktop Mn Max SoC paired with the "GPU-specific" SoC; resulting in a Mn Ultra SoC that is heavy on the GPU cores; then pair two of those for the GPU-heavy Mn Extreme variant...
Another variant use for these "GPU-specific" SoCs would be multiple dies on a PCIe card, for an Apple silicon GPGPU; stretching the R&D of these ASi GPGPUs could be achieved by also allowing them to work in an eGPU enclosure, for use with all & any ASi Mac that might need a GPU (compute & rendering) boost...
OK, so there are now 3 Max variants - Laptop (0 UF), Desktop (1 UF), Extreme (2 UF)?
OK, that part is understandable. You're just connecting an Mn Max to a Mn GPU via the existing UF connector.
Yes, but how? You've already used the UF connector for the GPU. Unless you are suggesting 3 variants of the Max, which would seem to cut into economies of scale (especially as the model with 0 UFs would be by far the most popular).
The studio goes up to 128gb ram . is that enough ? is that a limit for the cpu ? sounds like enough to me ....
128 gigs is probably not enough for the lifetime of the machine. Some people should be ok with that but... the 2019 Mac Pro goes up to 1.5 TB. I would guess there are plenty of users already running 256 gigs in their Intel Mac Pros.
IF there were two 'spare' , standard M.2 SSD slots on the logicboard then they don't really have to for the SSD pricing. Just skip theirs. That will work just fine vast majority of folks.
It is actually kind of goofy to use Apple SSD as a primary target for heavy duty scratch drive work. You are going to wear it out eventually. It is far better that is just a completely secondary volume completely detached from usage from the "home and apps" drive. At some point with high end workloads putting all of the data usages onto one single large capacity device makes less and less sense. The whole one, and only one, drive is good enough crumbles at higher workloads.
Inside a Mac Pro with multiple PCI-e slots a relatively very cheap M.2 caddy and the user can get to the M.2 drive anyway. The barrier to make that move is very low so it makes extremely little sense for Apple to be obnoxious about it . They are not 'stopping' anyone. It is just one annoying easy step around them.
If all the RAM DIMMs are gone on the back side of the logicboard, then dropping two M.2 sockets in that now empty place 'hurts' how? Even more so if take away the MPX x8 PCI-e lane provisioning for the TB controllers. Freed up lane assignments for that too. Assign two x4 PCI-e v4 and have something 'useful' to trade for the DIMM slot space. The RAM DIMMs super fans aren't going to be happy but at least tried to do something value add with the space.
[ If the 'spare' slots for extremely proprietary Apple SSD modules ( from the second , redundant SSD controller in a Ultra/Exreme . That is just plain greedy bozos. Bound to piss off even more folks. who are going to walk away with more money than they would gain with the gimmick.
There are about just as many 2023 user workstation motherboards on the market with zero M.2 slots on them as there are those with zero SATA sockets. It is almost the same situation as it was back in 2019 when Apple put a SATA socket on the motherboard. ]
The $/CPU core and $/GPU core prices. It isn't commodity consumer prices, but is relatively little volume for these dies also. It isn't going to be super cheap. Apple jumping off the shared high volume 'food chain' means don't get the same volume discounts on die work. Apple is making fat gross profits at the prices for the Ultra/Max etc. , but if they are going to stay competitive over the long term they have to fold that money back into staying in the game.
RAM. if they charged a bit less then they would likely sell more. Those are not completely standard LPDDR5 memory packages.
See end comment...
Yes...
I can only assume two UF connections would be needed for a 4-way SoC, which is what a Mn Extreme SoC would be...
Apparently this is the issue Apple is still figuring out, hence the "cancellation" (more likely just a delay) of the M2 Extreme SoC...?
It seems one of the main issues right now is getting more GPU horsepower into the high-end ASi SoCs, one solution may be the "hybrid" SoC, one "regular" SoC & one "GPU-specific" SoC...
Another solution may arrive with the 3nm process variants, allowing more cores to be jammed into the same square millimeters of the preceding larger processes...
It will probably be at least two or three generations to get an idea of what Apple intends in regards to more horsepower in the ASi Mac Pro, all of this is just my thoughts towards how Apple could handle these issues...
The M2 will easily bring this up higher, but not clear if it will be high enough for most pros. And even if were, making you decide up front with no chance to later upgrade is a bit of suck.
But Intel charged an extensive >1TB RAM tax for that upper limit. Apple probably didn't ask for that. They were more interested in slapping their Apple Tax on top of the Intel Tax and collecting larger revenue. There were cheaper Xeon W-3200 options that Apple completely by-passed on purpose that just happen to make them more money.
The Intel Mac Mini is still for sale over two years later because Apple isn't all that pressed about the Max RAM capacity feature at all. I highly doubt it is any different in the Mac Pro space where they have gotten even loss released.
AMD metal works that way on Intel. The MP 2013 compute card was always an GPU package exact twin of the display GPU. It is optionally run with the displayPort outputs not hooked up but the internals are the same as what is being run for display. If this mode is there why hasn't apple shipped it to Thunderbolt connected eGPUs for over two years? Or is it working and nobody is talking about it?
It doesn't have to be as a "production ready" driver. Just as an 'broad beta' test.
My problem with the "Apple is going to pop out of the woodwork with dGPU support on macOS on M-series" is that they are sure waiting a heck of a long time to due the extensively broad beta testing they'd need to get to have something that was 'good and stable' at a Mac Pro launch.
The A14Z developer transition kit didn't have Thunderbolt so it wasn't surprising why that didn't show up in year one. But the 'odometer' is past two years now... and still nothing... yet all the core foundational abstraction work is suppose to be just sitting there entirely ready. That was surprising (to me) at WWDC 2022.
A couple of years before APFS rollout out as a default it appeared as an optional file system a broad group of users could kick the tires on. Apple initially said they were going to roll out APFS as the default on Mac in a year ( which seemed like a joke or at best grossly naive. It didn't happen because there are way more Macs that have more than are not restricted to just one, and only one, Apple SSD only internal drive. Deploying a highly critical driver into a very broad ecosystem typically takes a substantive amount of lead time with real world workloads. )
It all depends on your use case. IMO a 128GB upper limit isn't sufficient for a pro level system. I have 512GB in mine.
This is the 2013 Mac Pro.
It's a failed paradigm.
No One wants a desktop computer that can have a second (third, fourth etc) GPU, which can't be used for display. The market doesn't exist. Stop trying to make Fetch happen. Non-display computer GPUs are a great option, but stop pretending they're the only use-case for extra GPUs, or that they're a large enough use-case to make a product viable, which has cut off an entire "more than 4 screens", and "more than mediocre viewport performance" use-case.
I would predict that the mac studio will get the M2 max, only, and the ultra m2 will be reserved for the Mac Pro in order to create some sense in the product line.
THere is also a chance Gurman's reports are not correct and they may surprise everyone with the chip, or specs, or more.
It works that way on any architecture, including Apple Silicon. It's not part of AMD Metal. It's part of just Metal.
Metal in no way requires a frame buffer. I have projects on Apple Silicon that use no framebuffer features.
Metal can also do graphics with no frame buffer or no attached display. Again - built into the base specification itself. I also have Metal projects that do graphics with no frame buffer - again on Apple Silicon as well.
From the ground up Metal was already built to be very agnostic and flexible. It does not depend on attached displays. It can even move frames from one card for render to another card for output as a basic feature of the core API.
The only AMD specific addition I know about is the Mac Pro supports moving any data (including graphics output) over the Infinity Link bridge. Normally Metal has to move data through the CPU first.
They did. From day 1.
Apparently no one is talking about it.
FWIW - This support goes way back before Metal too. I was doing dual GPU work on a 2009 MacBook Pro with a 9600/9400m (where one of those GPUs is headless) and pretty much any Mac GPU could be manually spun up for headless work even if no display was attached. Apple even did a graphics demo with dual headless cards on a Mac Pro back in like 2010 at WWDC.
Last edited:
Running with the Gurman info, the future for all Mac Pros actually looks pretty bright.
An AS Mac Pro that keeps the current design and simply allows the current and future MPX modules as accelerators for apps that would use them: FCPX, Resolve, Blender, and more would be great for both Intel and AS Mac Pros.
If the MPX modules simply carry over to the AS Mac Pro, then that is very good news and the best that could happen for both old and new Mac Pros.
EDIT: clarifying that multi-device-class would use CPU and GPU at the same time. For an AS Mac Pro, at least in Blender, this would effectively mean AS CPU + AS GPU + MPX GPU = all sources of power tasked at the same time.
An AS Mac Pro that keeps the current design and simply allows the current and future MPX modules as accelerators for apps that would use them: FCPX, Resolve, Blender, and more would be great for both Intel and AS Mac Pros.
- Base AS Mac Pro comes with empty slots. Stands on its own like the Max or Ultra and prices can be kept as low as possible. Just like the Mac Studio, for many users performance will be great in most scenarios just like this.
- Locked memory is a bit of a shame in a workstation. It's been a way to keep purchase prices in check and then upgrade later. Now, users will likely have to commit to a bit more than they need—especially when taking into account that selling a Mac Pro later, which is starved for memory, might make it a dud in the market. Let's hope that Apple recognizes this and keep memory upgrades reasonable.
- Adding MPX modules would solve the main weakness of the Apple Silicon design. You would end up with a very versatile computer with great single-core, multi-core, and GPU performance (upgradable). It won't change the fact that Nvidia owns quite a chunk of the GPU market, but still.
- Apple would keep updating the MPX modules which would add years of value to Intel Mac Pros. These Mac Pros also get to keep a small ace up their sleeve as the last modern Apple workstation that can dual boot into Windows, no problem.
- If Apple ship an AS Mac Pro with stand-alone GPUs, they should offer eGPUs for Mac Studio owners as well.
- With an AS+GPU official design, hopefully, future software development will further optimize the use of both sources of power simultaneously. In Blender, for example, you can choose CPU or GPU for rendering. In the development forums, there has been talk about allowing for "multi-device-class", meaning CPU+GPU, rendering as future development. If that comes to pass, suddenly upgrading the Intel CPU in older Mac Pros becomes attractive again. Finding cheap 28-core Xeons on eBay is classic Mac Pro territory. In DaVinci Resolve or FCPX, you could have the AS GPU cores drive displays and update the interface for butter smooth interactions, while the GPUs work in the background doing pure calculations.
If the MPX modules simply carry over to the AS Mac Pro, then that is very good news and the best that could happen for both old and new Mac Pros.
EDIT: clarifying that multi-device-class would use CPU and GPU at the same time. For an AS Mac Pro, at least in Blender, this would effectively mean AS CPU + AS GPU + MPX GPU = all sources of power tasked at the same time.
Last edited: