Not really, it's can top at 128gb/soc if available big enough dram modules (maybe even more), Mac studio on m2 ultra likely also to offer 256gb
Last edited:
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
If the trashcan resurrects, doesn't matter if having 4 M2 max, it will be just an fatter Mac studio, not an actual Mac Pro.
Mi guessing is apple readies an Mac Pro in the shape of an twice Mac studio with better cooling and dual m2 ultra, and addressing the GPU power issue by external tb4/whatever e-gpu.
It seems more obvious to me that we'll see what I have expected all along: A refresh of the Studios with the M2 Max and M2 Ultra, and a new Mac Pro pre-announced using M3, possibly in an "M3 Extreme" version, but more importantly with add-in compute accelerators, probably just GPU and ANE or both on the same card.
This lets Apple use their own GPUs, and now with hardware RT, and they'll appear first in the one machine that was originally capable of raytracing in hardware (putting aside the weak GPU's in some MBP's). This will be as much an announcement of their capabilities in the CPU and Mac Pro with expansion as it will be a launch of their new GPU capabilities.
This will also have a perfect tie-in to the VR/xR headsets and development for those.
As for a case, I don't see why they wouldn't reuse the practically brand-new design for the existing Mac Pro. No need to make it smaller and they can leverage all of the investment in designing and building that case. I see zero chances of Mago's fantasy return to the trashcan. It's possible there will be smaller versions of the current Pro chassis, but I really doubt that. Apple doesn't change external hardware form factors frequently, and they even stuck with the tried and true Mini design aesthetic for the Studio.
That's my final guess. I hope I'm right!
This lets Apple use their own GPUs, and now with hardware RT, and they'll appear first in the one machine that was originally capable of raytracing in hardware (putting aside the weak GPU's in some MBP's). This will be as much an announcement of their capabilities in the CPU and Mac Pro with expansion as it will be a launch of their new GPU capabilities.
This will also have a perfect tie-in to the VR/xR headsets and development for those.
As for a case, I don't see why they wouldn't reuse the practically brand-new design for the existing Mac Pro. No need to make it smaller and they can leverage all of the investment in designing and building that case. I see zero chances of Mago's fantasy return to the trashcan. It's possible there will be smaller versions of the current Pro chassis, but I really doubt that. Apple doesn't change external hardware form factors frequently, and they even stuck with the tried and true Mini design aesthetic for the Studio.
That's my final guess. I hope I'm right!
Right, he keeps saying "rumored", but literally, he's the only one who is still pushing that agenda. No sources provided (as usual) or any proof of anyone else on the net hinting that apple will revive the Trashcan.
It's solely a fantasy of his that he's been trying to push since apple filed those cheesegrater patents in 2021 (which were posted on this very website, and he's been trying to pull off like he knows something).
🤣
They have already blown past the two year timeline of both the WWDC 2020 and Fall (Nov) 2020 product info by saying nothing. Blew past a year after the "Mac Pro later" comment too; again NOTHING. Saying noting now would be substantively different how? (after WWDC 2023 is still in the nebulous "later" zone).
They could talk about it later in June , or July , August ... not going to get much worse than it already is.
If they had something that isn't 6 months off into the future why didn't they say something earlier in the year. They were ALREADY in the hole when the year started even if cherry pick the Nov 2020 as clock starting.
If the system is more than 6 months out then just pointing a finger at just how much deeper a hole they are digging; haven't even hit 'bottom' yet.
April 2017 hint at going to do something. April 2018 "not this year"
Apple has already set the prices for the M1 Ultra and it doesn't fit anywhere in near inside the $2999 price point even in the slightest. The M2 generation dies are even bigger . So Apple's costs , if anything , have gone up not down. Apple is going to throw their margins out the window and sell everything lower? You might try smoking something less strong. The shift away from Intel has never been about trying to lower the overall average system end user cost price point.
Extremely likely that when Apple cranked up the Mac Pro entry prices 100% in 2019 they had an very solid idea of what they wanted to charge for an Ultra (and higher). That price increase was just as much setting the transition stage as it was "undoing" the path the MP 2013 went down. Container , boxes with slots at the $2,999 range .. Apple isn't heading back to the 5,1 era.
Again off in looney tune land. The Studio is the iMac 27" replacement. "Queen is dead , long live the King" ... Studio has that slot now. It isn't going away. In fact, given recent leaks, more likely to be on stage at WWDC 2023 than the Mac Pro is. Apple talking about something they can sell in a couple of weeks versus .. hey look at this thing in the glass case that is "look but don't touch". What is a better use of WWDC 2023 stage time??????
If Apple is dropping the DIMM slots and 3rd party GPU drivers then cutting the number of slots does a whole lot of nothing good.
If down to just one or two slots then a Mac Studio inside of one of these
xMac Studio - SONNETTECH
3U enclosure to install and secure one Mac Studio with optional PCIe card expansion in a standard 19-inch rack.
is largely just covering the same spot. Six slots with real one , or two , x16 PCI-e v4 backhaul would be substantively different.
Adding PCI-e passthrough to the hypervisor/virtualization stack would make lost more sense to keeping 6+ slots also.
Also using the same box with extremely minor adjustments keeps the Mac Pro out of the "Industry Design" resource choke point. "Half size" just opens the door to "insert radical new design idea xyz that makes a 'statement' " into the mix. Mac Pro doesn't need any more delays at all. It is already 'late' and 'even later' is not going to help.
The only reasonable reason to go to 'half sized' would be that the PCI-e backhaul was some kludged Thunderbolt base that really wasn't any different than the Xmac Studio. And that just trying to handwave cut costs to complete with that. Which is more than kind of pointless.
I'm pretty sure even if there was an M2 Extreme (which the silicon itself doesn't support) they could probably get it into a Mac Studio case. There isn't really any need to do a new trash can.
Heck - most the reason for the trash can and the thermal core was the dual GPU. So even the "Itz gonna have a thermal core!!" thing doesn't make any sense. It's an SoC. There is only one chip. You don't need a thermal core. There's nothing to sandwich the thermal core in between.
Unless the XR headset SoCs also have hardware RT , that would be relatively pointless for dynamic content on the headset. Raytraced stuff plopped into a context where the lighting sources are completely different in reality helps you how? It likely isn't dominantly a VR headset where drag in static lighting from somewhere else.
If Apple is making the M2 Ultra exactly the same way they made the M1 Ultra then waiting for M3 is probably more so waiting for an actually decently good chiplet design rather than to continue to pound round pegs into square holes. Hardware RT may come in as a side-effect .
Errrr, probably not. Studio logic board from iFixit Teardown :
Mac Studio Teardown: No Upgradable Storage … Yet
We’ve got storage answers, and heat sink details, and a quick look at the Studio Display, this teardown is chock full of fun.
www.ifixit.com
That outer yellowish ring is likely the boundary for the Ultra package needs to fit inside of. There are still functionality that takes up very substantive amount of space that are not inside the SoC.
It is sizable folding puzzle to just get the Ultra's package and associated memory packages in there. Double the number of memory packages and that alone will blow the studio chassis out of the water. Let alone the additional compute die(s) and thermal overhead. ( Apple may shrink the compute dies over time (or keep same size and add 'more stuff' ), but the memory packages are not likely to get dramatically smaller at all. For LPDDR_ not on same fab die shrink path. ).
There is hocus pocus of just flip the logic board vertical and make Studio 'box" taller... still 'boat anchored' on the vacuum cleaner oriented, single fan.
Over time Apple can leverage TSMC N3 , N2 , etc to put more performance into the Studio case. But much bigger packages on far more less bleeding edge densities probably not so much.
Don't need a "new trash can" because already have that subset covered.
Not really. Apple was also massively ejecting anything other than one , and only one , internal drive. Even if had just a single GPU ( remove dual from the table) it still would not have 'worked' if keep the internal drive count the same ( 4 drives in MP 2009-2012 ). Baked into the MP 2012 was a lot of "SSDs make stripped spinning 3.5' drives obsolete" when looking for driver throughput performance.
That whole focus on a single internal drive thing was equally as big of a FUBAR as the mandatory dual GPUs. Probably bigger. Apple's hard push to make folks to buy just one , super expensive Apple drive is as misguided in 2023 as it was in 2013. Apple's APFS doesn't really want to deal much with optical or stripped SATA drives much anymore , but even multiple PCI-e SSD drives are not uncommon in the Mac Pro user base workloads. SAN/NAS storage usage is up , but not that high.
It makes sense if want to throw gasoline on the "I'm not happy with Apple's effort on Mac Pro" fire.
You forgot about the "Compute modules" thing he is also rattling on about. I have strong suspicion that the XR part of the WWDC keynote session will explain far more about what Compute module" is than the MP 2013 design constraints does.
Probably similar to the warping of leaks that got the Promise Pegasus R4i MPX storage module twisted into "next Mac Pro is going to be totally lego blocks assembly".
My thinking is they could maybe just build up. Stack a second board on top of the first - like a small version of the 2009 Mac Pros "CPU on a board" system. You'd loose some vertical space but probably a workable amount. Apple could even make the Mac Studio a tiny bit taller to compensate.
I think cooling would probably be fine. Apple might need a new fan curve - but the fans are still pretty conservative on the existing Mac Pro.
Also it's way easier than shipping an entirely different chassis.
Yeah. Compute modules make a lot more sense in a device that would need to have a series of connected sensors - and probably needs to deliver more processing power than what would be in a single SoC. Over time they could replicate that work to other products, but it doesn't sound like something they could just bolt onto M2.
The flaw in the 2013 Mac Pro was also that a thermal core requires the modules to be thermal pasted onto the core. That reduced any expandability to "Apple Store servicing." It feels like Apple learned that lesson with the 2019 Mac Pro. I can't see them repeating that again if they're going to have modules at all.
Last edited:
errr, stack it how. The MASSIVE heat sink on top of the first SoC package.
And "fans" is a pragmatically a misnomer here in that they are both blowing into the same heatsink. Thermal core coupling all over again!
I wish iFixit still did the picture treadowns but useful to look at the whole tearday video on the teardown page. However, this is about the timeline point that is most relevant to the "folding unrolling".
About 135 seconds in ( about 2:15 in)
The other part is where the power supply was on the initial bottom "layer cake" . Apple already had to resort to 'stacking' just to get the power supply in there for just one Ultra. If put another Ultra SoC worth of power in there would also need to stack in another power supply layer to go with the additional logic board stack.
And still just pragmatically one input vent , one heat sink , and one output vent. As long as they just have one input vent and one output vent then pretty much the thermal coupling problem is still there.
Has the same 'shared heat sink/pipe' path issues that the laptops have. So the "fan curve" isn't really going to solve the problem anymore than adjustments to the fan curve could have solved the MP 2013 issues either.
They aren't going to swap out someone's $6,000 Intel order for an Apple Silicon chip. You're off the rails a bit lol. 3 more days.
Called Just in Time manufacturing and inventory control.
Pick the most common configuration that Apple talked about 16 cores , W5700X , 48GB RAM , 512GB (or 2TB ) storage and watch the dates change.
The more least requested configuraiton the longer the ship time .... because ... shocker ... pragmatically they are making it after you order it and putting it on pre-packaged , cheapest shipment slots they have arranged.
XDR display... longish gap times also for basically its whole lifetime as a product.
Unless a new Mac Pro is the world’s best kept secret, it’s almost certainly not going to be released at WWDC. Now the entire range shares the same SoCs, though, it's even more tricky to preview future models. If the next Mac Pro uses an M3 Extreme, talking about it in any depth would reveal coming architectural enhancements. This would be a bummer for the M2 Max / Ultra Studio, likely to be revealed in the same presentation.
The Mac Pro could be discussed in general terms, but this would kill the buzz of a later full reveal. The information would have been dribbled out rather than given a big splash. OTOH, if Apple worry about high-end users defecting to Windows / Linux, which seemed to be the concern in 2017, perhaps in this case throwing them a bone takes precedence over 'buzz'.
And given we're firmly back to a pattern of long waits and no news, will Apple just leave the new MP to sit for years following release? History suggests so. At some point, the MBP and Studio will be on M5, with faster ST and the latest features (AV1, better RT etc.), and the MP will only have its higher MT performance and PCIe throughput. It's true that high-end iMacs often had higher ST performance than the Mac Pro in the past, but at least the latter had big / upgradable GPUs in its favour.
The Mac Pro could be discussed in general terms, but this would kill the buzz of a later full reveal. The information would have been dribbled out rather than given a big splash. OTOH, if Apple worry about high-end users defecting to Windows / Linux, which seemed to be the concern in 2017, perhaps in this case throwing them a bone takes precedence over 'buzz'.
And given we're firmly back to a pattern of long waits and no news, will Apple just leave the new MP to sit for years following release? History suggests so. At some point, the MBP and Studio will be on M5, with faster ST and the latest features (AV1, better RT etc.), and the MP will only have its higher MT performance and PCIe throughput. It's true that high-end iMacs often had higher ST performance than the Mac Pro in the past, but at least the latter had big / upgradable GPUs in its favour.
Last edited:
Mine took quite a long time to get it shipped from Apple, 16 core, 1TB storage and W6800X 32GB. I was desperate to get that machine as well. I looked at the maximum spec Studio once and gave it no further consideration. The 7,1 has proven its worth in operation.
The last 2 Mac Pros and the iMac Pro were all previewed at WWDC but only launched much later in the year. If there's a new Mac Pro worth writing home about, that makes sense.
Which is why I can't believe there is still a market for the Mac Pro. Since ~2012 the Mac Pro has spent most of its time as 3+ year old tech that definitely probably maybe might be discontinued and replaced with something radically different at the next Apple event. Mac Pros aren't impulse buys for "prosumers" (the 2019 version doubled down on that!) and pro customers have to plan workflows and budget projects around them.
If nothing concrete about the Mac Pro is said at WWDC I'd have thought that the Mac Pro will have to be declared "missing - presumed dead". If that's the way it's going to be, it's time for Apple to come out and say it.
As for M3/3nm - sure, its likely to be the most significant step forward in power since the introduction of M1 but I'm not clear how it solves the problem of making a 2019 MP successor with Apple Silicon. The M1 Ultra already matches the 2019 MP on CPU power and leaves any other integrated GPU in the dust - but Apple Silicon just isn't designed to run 1.5TB RAM or drive multiple high-end PCIe discrete GPUs - and squeezing a few extra cores for a few less watts thanks to 3nm isn't necessarily going to change that.
Meanwhile, something radical using multiple Mx-Ultra compute units would be fascinating but not necessarily useful for the customers who only stuck with Mac Pro because it was cheaper than adapting their workflow to use cheaper, more powerful commodity kit.
Yes, because Studios may not be the solution for 2019 MP users but they do what they do very well, offer a distinctive product, and are a much better application for Apple Silicon than trying to kludge it into a "me too" Xeon tower system equivalent when it really isn't the tool for the job.
Apple have been lukewarm about investing big money in developing Mac Pro in the past - the 2019 MP is just a Xeon-W tower with PCIe and AMD graphics, plus a few flourishes for routing power and Thunderbolt to PCIe slots and a MacOS dongle in the form of the T2 chip (developed for the money-making MacBook, anyway). Most of its headline features came from the new (in 2019) Xeon-W, It would be rather impressive if Apple suddenly poured huge resources into making a Xeon/Threadripper-killer version of Apple Silicon that was only going in their lowest-selling system.
I agree.
I'm thinking more in terms of a move to a full chiplet design, or perhaps additional UF connectors, rather than just the extra cores / speed bump 3nm would enable.
It's natural, but a lot of the speculation does seem to fixate on how Apple will exactly replicate their previous Xeon system with ASi. The fact is, they can't, and even if they could, it would probably cost twice what a Xeon / Threadripper equivalent does. Long term, no-one is going to pay that kind of premium just to use macOS over Windows, when they're spending all day in a small number of pro apps. Plus much of that market demands Nvidia GPUs, and that's not going to change.
Most of the value adds were luxury features, that while nice, don't really offset the benefits offered by Windows equivalents (better value at low end / more power at high end / CUDA).
Exactly, the chances are zero.
Except I don't quite get who the ultimate target market of the XR stuff is - VR gaming/simulations, tethered (or even WiFi'd) to a stack of local "compute modules" doesn't sound like Apple playing to their own strengths. Most of the talk seems to be about AR and for that the money will be in something that can be worn stand-alone or as an iPhone peripheral. WWDC may be aimed at developers, but they're probably going to want developers to target mobile apps.
It's not thermal paste stopping Macs from being upgradeable - applying paste isn't rocket science and the Trashcan was never designed to be upgradeable by non-tech people. The flaw in the 2013 Mac Pro was that it had a triangular core that assumed the heat output would be fairly evenly spread between a CPU board and two GPU boards (not 1, not 3... and 5 is right out...) and that model really didn't work out. Plus, it was dealing with Xeons and workstation-class AMD GPUs that put out a lot of heat - the Studio Ultra may look like it has a big heatsink but it's nothing compared to Intel/AMD systems (https://www.amazon.co.uk/Noctua-NH-D15-Premium-Cooler-NF-A15 - and that's just for the CPU).
For a Mac with multiple Mx Ultras it would make more sense to fit a bunch of them in a 19" Rack case, which is something likely to appeal to customers needing that sort of power anyway. You can already get a rack bracket for 2x studios - and if you just used the mainboards alongside a shared power supply you could greatly reduce the height, even more by making the units deeper and lower with the heatsink behind & connected via heatpipe.
I'll pose this question to the group... Just as the Mac Studio + Display is supposed to replace the iMac, would a Mac Studio plus egpu/externalPCIe enclosure... or "compute modules" .. be a decent Mac Pro proxy if you squint?
To me it just seems like the main Asi MP issue is GPU performance vs a Wintel HEDT box with a dGPU. Could apple solve that problem by basically running a TB4 cable to a powered box with a GPU in it? an M2 Mac Studio Ultra with a eGPU would be 192 GB ram, and a 7900XTX running at 80% bandwidth? I think that would cover all but very high end needs.
It just crossed my mind as to why the MP is so "secret".. if its because its just updates to I/O, in the M2 Mac Studio, something better than TB4 to interface with this external "compute module" we have heard about?
To me it just seems like the main Asi MP issue is GPU performance vs a Wintel HEDT box with a dGPU. Could apple solve that problem by basically running a TB4 cable to a powered box with a GPU in it? an M2 Mac Studio Ultra with a eGPU would be 192 GB ram, and a 7900XTX running at 80% bandwidth? I think that would cover all but very high end needs.
It just crossed my mind as to why the MP is so "secret".. if its because its just updates to I/O, in the M2 Mac Studio, something better than TB4 to interface with this external "compute module" we have heard about?
Last edited:
No, a TB4 PCIe expension chassis does not provide the bandwidth wanted by a number of add-in cards; (GP)GPUs, NVMe M.2 RAID, High-speed networking, 8K video I/O...
The basic issue with a lot of these ideas is that whilst it may solve Apple's problem of 'how do we cobble together a workstation from laptop SoCs?', it's hard to see the benefit to the user. If solutions like the above are either more powerful or lower cost than competing Windows solutions, then fine. If they're less powerful at the top end, it may still be fine, depending on the user's needs. If they cost more, it then becomes a question of how much more, and how much of an Apple tax one is prepared to pay to use macOS rather than Windows.
Whereas ASi's power efficiency and low heat output is arguably game-changing for laptops, and convenient for compact desktop computers, it's a 'nice to have' at best for a desktop tower. So Apple's nice hardware isn't the decisive factor in this situation, and it really comes down to price and the quality of the OS.
User-upgradable memory, storage, and display graphics are the minimum viable product for a Mac Pro. If it doesn’t do that, what’s the point of it existing as opposed to just buying a Mac Studio?
That isn't how it works. It suspect that Apple has some really secondary VR like things attached to the headset because there is a larger set of folks that have some 'Sci Fi' exposure to VR than AR. But like the iPhone didn't 'bet the farm' on gaming as a primary purpose for an iPhone , I really don't think it is likely that Apple's primary target here is VR at all.
Reportedly Apple nixed/killed the concept of running this as a tethered heads years ago. Ive and some others didn't want any wires (and wireless isn't all that tractable at higher resolutions) so the primary mode is to run the computations inside the headset.
Reports also indicate that there are two SoCs in the headset. When Apple has two GPUs inside the MP 2013, one was named the "Compute GPU" and the other the "Display GPU". if there are two SoCs inside the headset , then there is a more than decent chance they'll be given 'assignment' names in addition to some M-xx or X-xxx or whatever longitudinal product name Apple wants to slap on them to increment the numeric suffix on. They are probably not monomaniacally symmetric ( so in that sense different from MP's pair). One is probably specialized for sensor data fusion , screen handling , object inference , etc ( with a custom collection of fixed function units aimed at doing all of that a much lower Perf/Wat than general GPU/CPU/NPU cores can) . The other is probably running more of the basic iOS/iPadOS workload that generates video output that will be layered on top of reality. The video output here is perhaps run through the other out to the screens.
One of those gets tagged "Compute" module because it is doing the compute part that the other one doesn't do.
There was several threads a while ago where folks had taken some patent about "mGPU workload distribution" and turn that into a utter fantasy about discrete GPUs. When actually looked at the specific definition inside the patent could see ( if not delusional) that those are internal GPU core clusters. There are lots of folks who "want to see" a discrete card that they can plug in , so they desperately mutate anything close into that. I suspect something very similar is going on here and that these very high end HPC, compute cards are largely the result of the 'telephone game'.
These "compute modules" in some reports are running iOS ( or a close derivate of iOS). Some computational box that plugs into wall socket current to do "Nvidia killer" GPU compute workloads..... errr why would Apple throw iOS at that task?????
I suspect that "iPhone peripheral" will be a secondary option mode. Reportedly, there is a socket on the side where could sit and tether it to a Mac as an "external display" for a period of time. They could do the same thing with iPhone and also run 'notifications pushed to another device' like the Watch. I think those are somewhat placeholders until the xrOS software ecosystem gets fully populated with native stuff has opposed to 'quick hack' iPadOS ports (that probably do more to blow up battery life , than are useful in motivating a $3K price point. )
I suspect it is a mistake to think this thing is like what the Apple Watch was when it came out. A device that was woefully depending upon something else to do anything 'useful'. Lots of things point to Apple putting a 'game changing' amount of 'horsepower' inside the headest. Way , way , way more than anyone else has done so far.
And that will bring change. Change in the preconceptions of what can and cannot be done on the device. Has happened in PCs ( in 80's still heard Mainframe , Minicompute folks talking about how need a 'real computer' to do blah blah blah). Happened with SSDs wiping out huge chunks of the HDD market. Happened with the smarthphone.
Almost every other headset vendor out there is buying off the shelf chips from Qualcomm and others to power the headset. Who has done a large completely custom SoC for one ( at somewhat reasonable prices) ? [ Microsoft has coprocessor for Hololens but not the majority of all the computation required. ]
If the rest of the Mac line up was not there and Apple only had a $6K Mac Pro that was the whole line up and tried to pitch that machine to folks who primarily just need to read email , surf the internet , and play a casual game they probably wouldn't 'get it' . Whereas if you go into a extremely not mainstream shop that does large scale video editing the Mac Pro isn't that hard to sell at all. That's the problem at this point. There is huge faction of folks who think Apple is 'talking' to them and they really aren't. Apple will take you money if want to buy one just because you can... but that isn't the point.
This headset should be aimed more at someone who has to fix/troubleshoot a $20M jet engine , than virtually try on $20 dresses in a virtual world. At $20M/year pro athlete workouts, than a housewife looking for a yoga session on Apple Fitness+ . A surgeon doing AR enhanced surgery , than a parent looking for tele-medicine help for their baby with a fever.
It is a productivity enhancement device more than "escape from reality" device.
It would be simpler to just do a "mac on a card". If each board is running a independent macOS instance then just follow the normal PCI-e standard card practice of putting the power supply off card. Get rid of most of the external ports to cut down on non SoC components to the board (and other power drains). [ e.g., on EtherNet port , one TB port (video / DFU/ etc) and one USB C port (run KVM if want to). And software Ethernet via PCI-e back to host backplane/bus. ]
Apple wouldn't have to do the "rack chassis". They could just fit into other boxes ( like the Mac Pro) or any other 3U-5U box that took a full lenght , full height card ( all the 400-700W GPU cards popping up means those will be around even in non OAM card from. )
Effectively it would still be a Mac. Just one that wasn't completely enclosed inside of eco-friendly, recycled aluminum. ( not a "compute module" but a complete Mac system SoC , SSD , ports , fan(s) etc. )
Could do a fanless one but just recycling the Afterburner card and replacing a Mn Pro (?) (and SSD and few ports) for the FPGA if all that fit inside of a 75W budget.
Last edited: