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 All Models 40 core ASI with PCI-E slot is around the corner!!
- Thread starter Amethyst
- Start date
- Sort by reaction score
We
We may not see any 3rd party gpu support but Apple could come up with its own Gpus.
Apple is not willing to build a custom CPU for the Mac Pro. I would guess they're not going to do custom GPUs either based on that.
If they're doing GPUs, it'll likely be third party Radeons. I haven't seen any evidence that can field something competitive with the Radeon 7900 right now anyway.
I also prefer Apple not to do it because Apple is very famous with not properly supporting something they create like After Burner. Instead, the AMD drivers permitting us to throw some in would also suffice.
Doubtful...
A strong possibility...
Until we actually see the ASi Mac Pro, we have zero idea as to what will power them...
I am holding out for a custom SoC, just for the ASi Mac Pro, and probably also the Mn Ultra Mac Studio, to spread the costs over two platforms...
One would assume the ASi Mac Pro development is under super tight wraps, and will surprise us all...
If by "support" you mean release a new/upgraded Afterburner card, why would they...?
The current Afterburner is $2000, a base M1 Max Mac Studio is also $2000; anyone using the Afterburner card is using it for video editing, and would be better served buying said M1 Max Mac Studio for this task than in forking out a couple more grand (and placing the "old" afterburner card in a drawer) for a new Afterburner card...
Apple is done investing in the infrastructure of the 2019 Intel Mac Pro, the future is with Apple silicon...
Gurman says custom SoC for Mac Pro is dead. I tend to believe him.
And if Apple's not willing to do a Mac Pro specific SoC - I don't think they're going to do Mac Pro specific GPUs either. And they'd need to do something _really_ custom to try to fit in with unified memory on an MPX GPU.
By support, not only mean upgrades with more powerful ones but also road maps and some communication for the pros. The Trashcan Mac Pro just stayed there for 7 years without informing those people counting on Apple for their piepelines. Not only hardware but also on the software side, Apple always has a gap with the Pros. Besides, I could count so many softwares Apple abandoned and killed without proper communication with the pros. They killed Shake, Color, Aperture...I remember the times I started learning Shake. And then booom, Gone. I started using Color with my workflow; I used it on some professional works back in 2007-8ish and then boom it is gone. I was using Aperture for my own personal photos, then boom. Maybe the next one FCPX and Motion. Who knows??
Actually not. His rumors never included a Mac Pro that exclusively ran on the 4-chiplet solution. Some new formulation of the Ultra was mixed in there. the 4-chiplet SoC was just the SoC that was apparently exclusive to the Mac Pro, but it wasn't the only one. ( no real rumor of the "Extreme" going into some kind of iMac Pro+ chassis. )
An Ultra with an additional chiplet in the package to provision two x16 PCI-e v4 clusters would work just fine for the new Mac Pro . That is enough backhaul for six slots. ( Apple does 6 slots in the MP 2019 with two x16 PCI-e v3 clusters hooked to a dual input Plex PCI-e switch. Just do the same thing and would have the same 6 slots. Could just 'kill off' Slot 1 and Slot 3 being provisioned straight off the CPU package and the MPX connector (and provisioning) ). A substantial amount of complexity driving the board BOM costs up would disappear. They could swap that for folks paying for a more expensive entry SoC. ( an Ultra with mods. that is incrementally more expensive to make).
Apple doesn't need four laptop optimized Max dies to solve the I/O backhaul problem. The M1 Max has four x1 PCI-e v4 lanes. Four times that ( 16 x1 lanes. ) is still relatively nothing in terms of real, useful workstation backhaul. The M2 Max being used in the new laptops probably has the same x1 lanes. ( because the laptops don't need anything more ).
This could be done relatively straightforward with the laptop max with a relatively thin, 'shim' desktop I/O put in-between the two laptop Max's.
Turned sideways.
M2 Max [UltraFusion] very short PCI controller die that does lots of pass through mesh [ UltrafFsion] M2 Max
Don't want to separate the GPU clusters too much, but something thin shouldn't be that much of an increase in distance. That is closer to making a new package than a new CPU-GPU die. It doesn't interfere with the RAM layout. It would cost more.
Other low incremental cost is not to use exact twins and strip some of the Thunderbolt and redunandt Security Enclave , SSD controller etc off the "top" of a 'Desktop Max'. If 95+ % of the whole chip floorplan layout is exactly the same then doing another would be about the same as doing M1 Pro and M1 Max ( they too actually share a hefy chunk of common layout and building block components. Instead of 'chopping' the additional GPU cluster of the Max , just 'chop and replace' a portion of the I/O elements at the 'top' of the Max die. )
Both of those are way , way , way less than doing a discrete GPU. There is almost zero additional driver/firmware work to do. ( have a basic x1 PCI-e v4 firmware. Incrementally more to do whatever x16 needs). Removing Secure ELement and 'extra' TB controllers isn't going to cost much 'new' software. Even if Apple threw in a small SATA controller they could buy that IP off the shelf that already has a working driver. Or just toss in x4 PCI-e v4 to farm out to a discrete SATA and USB controllers.
The rumored one-slot-wonder could be provisioned in a cheesy fashion with an 'extra' TB controllers. But if the six slot report is right. ( which would justify the using an 8 slot case on the cheap), then they need something different. I think the problem was more so scaling up to 4 'chiplets' if they were not very effective chiplets.
Apple has already done two. So two wasn't a huge hurdle.
It is more than a bit of a oxymoron to "surprise" folks with drivers almost nobody has seen and to also have broadly stable GPU drivers. Even look at the 7900's roll out.
If all that Apple has is the N5P building blocks of the M2 Max then they really don't have anything 7900 broad scale competitive. Even if the "Extreme" wasn't too expensive to green light for high production.
The more affordable option for Apple is to just not to complete there.
Apple has not way of connecting a discrete GPU back to the host SoC with the elements they have included so far in their SoCs. The four x1 PCI-e v4 lanes of the Max aren't going to do it. If the dGPU is offpackage then UltraFusion is relatively useless. There is a pretty tall hurdle they'd have to get over just to connect back. If it is just via generic PCI-e v4 then there isn't alot of upside they are going to have over a generic reference design when it comes to costs. AMD is selling 10's of thousands per quarter without Apple's 'help'. So Economies of scale area already there.
Apple doing a GPU card that only boots inside a Mac Pro would be a extremely narrow limitation. that is worse broad market penetration than Intel's dGPU.
More likely Apple is going to do something incremental and affordable (to them) to address the issue.
[ I still think probably not going to get a display GPU drivers out of it even if some AMD drivers come. Although apple just now doing HDMI 2.1 and significantly behind the curve on AV1 and DisplayPort 2.1 is in about as non-compete as their top level, scalable-in-the-box GPGPU TFLOPs ]
Afterburner only worked on Apple's proprietary codec ProRes. Why folks couldn't see that was going to get folding into the SoCs after all the 'bugs' were out I have no clue. It is an Apple thing that only ran on macOS ... why would apple keep it separate long term???????
Once the iPhone SoC showed up with partial ProRes support that roadmap should have been very clear. Weaved into the SoC was where it was going. That is more of a "not really observing and listening" rather than "Apple isn't communicating a road map". Missing that is more myopic belief that modularity solves everything.
a M2 Ultra chip that has the same I/O bandwidth out as the M1 Ultra chip? It isn't strickly impossible , but it would be a dumb move. There is not enough general I/O out. ( two four x1 PCI-e v4 lanes complexes isn't going to make the cut with most workstation users. ). Apple could do some cheesy hackery for 1-2 slots with a extra Thunderbolt controller and in internal discrete Thunderbolt peripheral control to break out the x4 PCI-e v3 into a switch. That backhaul there would be pretty lame though, if practical/possible to implement.
Doesn't the M2 Ultra "have to have" the laptop I/O limitations? No. It is enitrely possible that they do something different for the M2 Ultra. Either leverage UltraFusion to put additional PCI-e output onto another chiplet in the package. Or simply just use identical 'twin' Maxes to form an Ultra ( one of the two dies swaps out some Thunderbolt and redundant other elements like SSD controller and Secure Element for more complete, wider PCI-e v4 I/O (e.g., x16 v4 or two ).
Neither of those require any radical or huge resign from the M2 Max they are using in laptops.
Probably not going to do some very large PCI-e lane provisioning (double the MP 2019's x64 PCI-e v3 bandwidth). Or chase after PCI-e v5 (v6) in the M2 generation.
Also during the introduction of the M2 Max today Apple's photo did not have a UltraFusion connector on it. The M1 Max that was in the side-by-side die shot had the UltraFusion connector re-Photoshopped out again. ( after showed the full Max die shot in the Ultra Introduction). The M2 Max is substantively bigger than the M1 Max. So there is a decent chance apple swapped the UltraFusion die space for more internal logic on the Max die. And that there could easily be another more "desktop" baseline chiplet for Mac Studio and Mac Pro that is incrementally different. ( e.g., shave off the top I/O edge and attach with UltraFusion so can sub in appropriate mixes of Thunderbolt/PCI-e v4 as convenient for Studio or Mac Pro deployment. )
The problem with M2 Ultra's that have M1 Ultra's I/O is that there is no creditable channel to hook more than one of those SoC together in a tightly coupled fashion. If all have is a couple of x1 PCI-e v4 slots there is not enough bandwidth to do anything useful. Could run a defacto 10Gb/E like link between the cluster nodes, but are most folks going to be happy with that?
Ultrafusion doesn't work over inter-card distances.
So back to needing something like x16 PCI-e v4 (at least ) for a inter SoC cluster cluster interconnect and a different approach to generation high end I/O than the M1 Ultra.
No. A M2 Ultra with two x16 PCI-e v4 clusters could feed a Plex PCI-e switch the same way the current Mac Pro 2019 does to provision out 6 slots. 6 slots , 8 slots not that much difference. Even more so if Apple sells the Intel and AS models side-by-side for 12-18 months like the Mini ( late 2020 - early 2023 ).
Storage cards , Audio capture cards , Video capture cards , Networking cards , etc ... they all take space. It in no way "has to be" Apple internal cluster cards. Doesn't really have to be any MPX connector bays in their either to use the exact same case.
Siamese SoCs, chiplets, whatever; Apple could decide to design something specifically for the ASi Mac Pro...
Maybe there is no UltraFusion on the new M2 Max SoC because those SoCs are intended for the 14" & 16" M2 Max MacBook Pros and the M2 Max Mac Studio...
Apple puts a good deal of focus on the ASi Mac Pro, with plans to introduce at WWDC 2023...
The first gen ASi Mac Pro also debuts the M3 Ultra and M3 Extreme "SoCs" (these my be SoCs, these may be chiplets...?) with hardware ray-tracing and up to 1TB of LPDDR5 SDRAM, but who knows how they will handle the ECC issue...?
If they kept the 2019 Intel Mac Pro in the line-up until they released the "high-end" ASi Mac Pro, then maybe they also release AMD drivers for the 7000-series GPUs, MAYBE they actually sell MXP W7000-series cards as well, and possibly a "spec shuffle" where the next tier up of resources (CPU cores, RAM, SSD) suddenly has the pricing of the tier below, with the bottom tier no longer an option...
Maybe they keep the MPX slots, but without any TB fold-back (or whatever it is), for use by ASi GPGPUs...
So the bottom four slots becomes two slots, no more covered up slots when using MPX modules...
Now about that new SoC in 2025 on the N3X process, with second gen hardware ray-tracing...!
Oh, wait, this isn't the "
Another reason it's not likely to be some hidden big surprise: a hidden surprise means no software packages will be ready for it. Apple would be shipping yet another Mac developers need to catch up to, with its own exotic graphics architecture. They'll launch with little to no software packages ready.
However - supporting Radeons would deal with that. Was your software package already optimized for multi GPU Radeons and Infinity Fabric on the existing 7,1? It will run just as well on an 8,1 with Radeon MPX modules. Was your software optimized for Apple Silicon? That will run optimally too using the SoC GPU. Was your software optimized for both architectures? Then it can use Apple Silicon and Radeons simultaneously for what each architecture is good at.
For a backup plan, it would be a pretty darn good product.
I have some information regard yesterday launch!
- MBP 14/16 is ready to launch since last OCT 22.
- Although there are pro chip Mac mini prototype has flown around Apple mothership last year, <<At OCT22 THERE ARE NO M2 PRO MINI in production release plan>>, the plan have been change after that. And the team has miraculously work to bring it production machine in such short time.
<<My Opinion: As I'm Mac Studio fan (my main machine is M1 Max studio) I'm so worry about Mac Studio future>>
- 24 inch M2 iMac is already pass software testing process and can be release anytime.
- Regard to Mac Pro .... << I will release info next week>>
.. haha just kidding, the Mac Pro is still in work although there are no new prototype come to my pal, he tell me that in software-wise the stability of this system is increasing significantly.
- I try to squeeze some info about 3rd party GPU, he doesn't tell me anything just reply me with
-like-emoji.
- he believed that if nothing change spec-wise (Mac Pro use M2 architecture etc....) we will get Mac Pro within THIS year not next year as some reporter report.
- MBP 14/16 is ready to launch since last OCT 22.
- Although there are pro chip Mac mini prototype has flown around Apple mothership last year, <<At OCT22 THERE ARE NO M2 PRO MINI in production release plan>>, the plan have been change after that. And the team has miraculously work to bring it production machine in such short time.
<<My Opinion: As I'm Mac Studio fan (my main machine is M1 Max studio) I'm so worry about Mac Studio future>>
- 24 inch M2 iMac is already pass software testing process and can be release anytime.
- Regard to Mac Pro .... << I will release info next week>>
.. haha just kidding, the Mac Pro is still in work although there are no new prototype come to my pal, he tell me that in software-wise the stability of this system is increasing significantly.
- I try to squeeze some info about 3rd party GPU, he doesn't tell me anything just reply me with
-like-emoji.- he believed that if nothing change spec-wise (Mac Pro use M2 architecture etc....) we will get Mac Pro within THIS year not next year as some reporter report.
Last edited:
Miraculously? Granted the shape of the logic board is different, but the component difference between a MBP 14" M2 Pro and a Mini M2 Pro is what?
Same case that have been using for over 4 years. Moonshot complexity how?
Recycled exterior backplane from Intel Mini ( sockets in the exact same positions )
HDMI socket. on both. ( associated software/firmware drivers should be done with the MBP 14" M2 Pro)
3 Thunderbolt sockets versus 4 Thunderbolt sockets ( when 4 TB sockets are on the current Intel Mini 2018).
New Wifi 6E and bluetooth. Extremely likely exact same discrete controller in both systems and M2 Mini. [ 'new' work where? ]
Ethernet 1GbE but probably extra x1 PCI-e v4 lane coming out of M2 Pro anyway. 10GbE controller probably hooks up same way does on M2 Mini. (So that manly done). [ have M2 era drivers for this stuff.]
SD Card slot gone so.... how hard to drop something for a board that never had it?
Same RAM limitations as MBP 14".
Likely can lift power supply from Intel model almost exactly. (or at least start from known tweaks to the M2 model. )
SSD NAND placement might be slightly different , but firmware for 4-8TB SSd .. done with laptop.
They have some board trace refactoring to do. Some thermal cooling system work (or tweaks). Firmware tweaks for configuration. But major, "stop stopper" re-design work?
There is some missing stuff from the MBP 14" but the M2 Mini covers most of that. And if that was was completely done .
They didn't have to wait until the MPB 14"/16" were in 'ready to ship' status to start to do initial prep work (which wouldn't need much to be "shipped around Apple". )
Honestly, not having a Mini M2 Pro was kind of a really odd move. Who was asleep at the wheel there????????? apple had said they were done with the Intel transition on the Mini but it was as obvious as a turd in a punch bowl that they were not. Who was drinking too much Cupertino Kool-aid?
If Apple's 'attachment' rates to the Studio Display with the Mac Studio are in the 30-40% range I wouldn't worry much. The average total system sales price would be up from the iMac. Apple walking away from higher prices is not very likely. If the attachment rate is 10% then I'd be a little more worried.
If Apple needs a headless model to drive up M2 Pro sales volume, then they also need a headless model to drive up M2 Max sales volume.
If the Mac Pro keeps it's $5,999 starting point then the Ultra Studio will be needed to drive up M2 Ultra sales. the Mac Pro is way too low to solely sustain a 100% custom SoC all by itself. Let alone two of them.
Whatever chiplet disaggration refactor Apple does to make the Mac Pro SoCs more affordable (for Apple) probably is not going to hurt the Mac Studio much. Likely still going to end up with a way to at least make Ultras for the Mac Studio. Maybe incrementally more expensive, but likely not too painfully so. Not enough to kill that 'half' of the sub-product line up.
If Apple brings back an iMac but models it on the iPad-on-a-stick mindset , then I doubt it would cover the space the Studio covers. Probably has more overlap with the M2 Pro Mini than the Studio. Which is fine. There is more user diversity in the sub $2K range anyway to support that. ( one reason why the MBA and MBP 13" still work well. )
The only part of Mac Studio that is likely a problem is the $1,999 entry price. I suspect that rises a bit to ease the price gap stress between the Mx Pro Mini and Mx Max Studio. At $5,999 the Mac Pro will be way bigger gapped off than the M2 Pro half of the Studio.
That actually makes more sense than Gurman's "waiting for M3" notion he keeps trotting out.
Maybe a Mn Extreme Mac Pro Cube to drive up Mn Extreme sales volume...?!? ;^p
It's miraculously because there are very short time for testing hardware and software to pass production stage, there are very short time to prepared production line and very short time to prepared marketing asset, thats all.
Apple can use some Draconian NDA's to prep a small/little number of software packages. If they got just Blackmagic Da Vinci Resolve working in multiple GPU compute offload , and 2-3 other packages that would like work to have something at launch while pointing to "slow motion" shops like Adobe as "going to start looking at it" candidates.
I wouldn't use it for production work, but folks could kick the tires.
I have doubts Apple is going to try to paint a picture that this 3rd party GPU path is the one they want most folks to use. The smaller number of software packages the more folks are focused over short term on software that is optimized for the Apple iGPU.
Short term that looks like a 'quick fix' but if those two software gets calcified over the longer term onto two diverging paths that actually turns into a problem. If have software which only works on 'corner case' Mac Pro configurations then long term economics are not going to look good for most of those packages more and more decoupled from the full mac software ecosystem.
If all Apple does is bring over drivers from the 6000 series MPX modules ( or just the upper end ones because that is where the biggest gaps are) then not sure going to keep lots of user base if this is just 'safe haven' for MPX 2021-2022 MPX sunk costs over the long term.
LPDDR5X SDRAM is pin-compatible with LPDDR5, so physically a drop-in swap...
Faster speeds, lower power, and higher capacity; up to 1TB in a 4-way SoC configuration once Samsung starts pumping out 64GB chips...
Should be a no-brainer for Apple to use in the first-gen ASi Mac Pro...
Faster speeds, lower power, and higher capacity; up to 1TB in a 4-way SoC configuration once Samsung starts pumping out 64GB chips...
Should be a no-brainer for Apple to use in the first-gen ASi Mac Pro...
So Mac Pro is still coming at a March event??I have some information regard yesterday launch!
- MBP 14/16 is ready to launch since last OCT 22.
- Although there are pro chip Mac mini prototype has flown around Apple mothership last year, <<At OCT22 THERE ARE NO M2 PRO MINI in production release plan>>, the plan have been change after that. And the team has miraculously work to bring it production machine in such short time.
<<My Opinion: As I'm Mac Studio fan (my main machine is M1 Max studio) I'm so worry about Mac Studio future>>
- 24 inch M2 iMac is already pass software testing process and can be release anytime.
- Regard to Mac Pro .... << I will release info next week>>
.. haha just kidding, the Mac Pro is still in work although there are no new prototype come to my pal, he tell me that in software-wise the stability of this system is increasing significantly.
- I try to squeeze some info about 3rd party GPU, he doesn't tell me anything just reply me with
- he believed that if nothing change spec-wise (Mac Pro use M2 architecture etc....) we will get Mac Pro within THIS year not next year as some reporter report.
@Amethyst
Might we get some clarification as towards the six PCIe slots in the latest ASi Mac Pro prototype chassis; specifically, are any of them of the MPX variant...?
Thanks...!
EDIT TO ADD: For the clarification to others, I ask with thoughts toward the possibility of ASi GPGPUs, not towards the return of AMD (or, as some wish, Nvidia) GPUs in the ASi Mac Pro...
Might we get some clarification as towards the six PCIe slots in the latest ASi Mac Pro prototype chassis; specifically, are any of them of the MPX variant...?
Thanks...!
EDIT TO ADD: For the clarification to others, I ask with thoughts toward the possibility of ASi GPGPUs, not towards the return of AMD (or, as some wish, Nvidia) GPUs in the ASi Mac Pro...
Last edited:
What would a pure GPGPU card have any signficant need need the MPX connector for? The primary function of the MPX connector is to provision Thunderbolt. a solely focused GPGPU would have no Thunderbolt function. So the need is what.
If the supplementary connector so 100% solely to provide power to the card ( so a 'no messy power cord' ) solution then it really not an MPX connector as don't need the bulk of the pins present in a MPX connector. Even with MPX Apple still put AUX power on the mainboard so still likely would be there anyway.
The primary mechanism in MPX bay for data transfer is over the primary PCI-e standard socket. If has a x16 PCIe -v4 edge on the card then what is so special?
P.S. Already stated in an earlier report that a AMD graphics card was plugged in , power uped and not recogniczed past being a device on the PCI-e bus (with no drivers). So more than likely a AUX power port is there (unless a huge effort to snake some power cable into the box from the outside. )
Last edited:
One reason for MPX is the power delivery, because Apple does not like cables cluttering their interior...
The "AUX power" ports are there for folks who want to put lower cost third-party AIB GPUs in their Mac Pro, rather than pony up for the official certified Apple MPX GPUs; and Apple will sell you the Belkin cable kit to use those "AUX power" ports...
So if not a full-blown MPX slot, then a shorter slot just for the power delivery, no need for possibly unused "AUX power" ports then, because the chassis will only support ASi MPX-Lite GPGPUs...
I think when we say „mpx“ card we just refer to the general concept of these bottom slots, not necessarily the technical details of the pin connector.
If - big if - there will be support for GPUs, I’m convinced it will be compute only and no display support at all. So no need for Thunderbolt connections.