Why not the debut of the M2? Although it would be weird to debut with the M2 ultra for the pro right away unless they also debut say the new MacBook Air with the base M2?
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 Apple Silicon Mac Pro Specs & Pricing?
- Thread starter Chevron
- Start date
- Sort by reaction score
Why not the debut of the M2? Although it would be weird to debut with the M2 ultra for the pro right away unless they also debut say the new MacBook Air with the base M2?
That’s what I said, M2 something in the new Mac Pro. Weird, but not forbidden by anything.
If Apple has M2 Ultra/Extreme SoCs ready, maybe they go nuts, everything refreshed (with a few redesigns as well) with the M2 family of SoCs before the end of 2022, transition complete AND on gen2 within the two year timeframe...?!?
Apple said M1 Ultra completes the M1 family. So it should not surprise anyone that M2 is on the horizon. On the other hand, it will be very surprising if the introduction of M2 starts from a Mac Pro with potentially the highest-end M2. But other than this, I don’t see many other ways for an Apple Silicon Mac Pro to have more than 128 GB of RAM. The 128 GB limit of M1 Ultra just doesn’t seem “pro” enough to me.
I guess we can only wait and see. If sometime in the summer or early fall an M2 MacBook (or Pro) appears, then a Mac Pro with the highest-end M2 in the end of the year would appear much more natural.
I guess we can only wait and see. If sometime in the summer or early fall an M2 MacBook (or Pro) appears, then a Mac Pro with the highest-end M2 in the end of the year would appear much more natural.
Apple Silicon Mac Pro Specs & Pricing?
Assuming the ASi Mac Pro will come with dual 10Gb Ethernet as standard, a NVMe RAID card would be in the most need of bandwidth... Audio cards (Pro Tools HDX) are good with x4 slots...? One PCIe Gen4 (maybe Gen5...?) x16 slot & three x4 slots for the 2022 ASi Mac Pro...
forums.macrumors.com
There is another possibility that, in retrospect, is more obvious. Two physically separate Ultra chips.
I have thought about the possibility of 2 x Ultra Mac Pro too. It's just the Apple wording is kind of a little bit against it.
Already mentioned in numerous threads: They said something comparable to that this is the last M1 series chip, it completes the series or something like that.
Another one was about programming: They touted that you can program just like it was one huge chip. No need to take into consideration that there are two chips in there. It would suggest that this is going to go on to the future, simplicity with programming fot ASi.
Ok, they could double up with the trick again with their Ultra Fusion. But I think it should then be called M1 "xxxxx" chip. So in that case that would not have been the last M1 as they said.
And as mentioned earlier by some users, it might become problematic with physical memory connections besides the chip(s). And memory should really go bigger than that of a Studio, for a Pro machine it should I think.
The fact they mentioned Mac Pro in the event, teased with it, might mean it's coming a little sooner than we thought.
THe price though, hmm. Some good aproximates above already, and going up..
Already mentioned in numerous threads: They said something comparable to that this is the last M1 series chip, it completes the series or something like that.
Another one was about programming: They touted that you can program just like it was one huge chip. No need to take into consideration that there are two chips in there. It would suggest that this is going to go on to the future, simplicity with programming fot ASi.
Ok, they could double up with the trick again with their Ultra Fusion. But I think it should then be called M1 "xxxxx" chip. So in that case that would not have been the last M1 as they said.
And as mentioned earlier by some users, it might become problematic with physical memory connections besides the chip(s). And memory should really go bigger than that of a Studio, for a Pro machine it should I think.
The fact they mentioned Mac Pro in the event, teased with it, might mean it's coming a little sooner than we thought.
THe price though, hmm. Some good aproximates above already, and going up..
Maybe it would work that way. It would get us only 128GB RAM though, until the RAM chip density goes up. Maybe we could get 256GB RAM for Mac Pro with that scenario, doubling the RAM chip density. That would be enough for me, but not to everybody. And what about the cost of 256GB unified double density RAM - well..
Really I don't have any idea. Some users have suggested it many times here on forums already.
Maybe it's a great idea. Or maybe it would add unnecessary complexity without real benefit. It would add another layer to existing data handling there is now.
I have read apple heavily uses flash memory for swap, like an extonsin of ram, already today. More with M1 than on intel, because of restricted ram sizes on M1. So it might give some speed advantages - or maybe not, because of layering too much traffic.
But yet again, interesting times we are living. I mean still we are, even if I myself can't jump to ASi for now, at least not yet.
Maybe it's a great idea. Or maybe it would add unnecessary complexity without real benefit. It would add another layer to existing data handling there is now.
I have read apple heavily uses flash memory for swap, like an extonsin of ram, already today. More with M1 than on intel, because of restricted ram sizes on M1. So it might give some speed advantages - or maybe not, because of layering too much traffic.
But yet again, interesting times we are living. I mean still we are, even if I myself can't jump to ASi for now, at least not yet.
Guess 2 makes best sense for a Mac Pro but adding Studios in Rack format sound like a supercomputer type of setup - that would be cool. Can MacOS handle separate ultras as one and even more important does the software needs to be rewritten? Even now it seem that the Ultra is not going on full speed in some software.
They have that kind of supercomputers. But I think the infra of it is way beyond 1Gb or even 10Gb ethernet. And beyond 40Gb thunderbolt too. And it would cost too to link computers that way and efficiently enough.
Example of self tests:
There would be some benefits at first growing the farm, but you would need dedicated software support. Very rare to find that support on software. I have tried C4D distributed renders. It was good and growing with performance, but just up to some point, like 5 machines or so. After that, it's diminishing returns. That experiment was with CPU rendering. It was called net render at first, it work through internet too, but it was unefficient, so it's was changed, updated, and it was called team render. It transformed to a local network renderer. That's how I remember it, maybe some C4D hard dies could elaborate and correct. I know a lot of people have used team render as a professional means. Maybe not today that much anymore.
As of today it's GPU:
This distribution of workload today would mean GPU rendering. GPUs are supreme for some work. There are GPU clouds already in function. I doubt it you could do this with some Mac Studio hardware without some 3rd party dedicated software support. And I believe it's the same with yet to be announced ASi Mac Pro.
Apples CPU oriented xGrid has not been resurrected, is it? And with GPU capabilities in mind this time?
That would be a surprise. At least in light of those -comments I've earlier mentioned. "You can handle it like one processor only, no worries about complexity".
Example of self tests:
There would be some benefits at first growing the farm, but you would need dedicated software support. Very rare to find that support on software. I have tried C4D distributed renders. It was good and growing with performance, but just up to some point, like 5 machines or so. After that, it's diminishing returns. That experiment was with CPU rendering. It was called net render at first, it work through internet too, but it was unefficient, so it's was changed, updated, and it was called team render. It transformed to a local network renderer. That's how I remember it, maybe some C4D hard dies could elaborate and correct. I know a lot of people have used team render as a professional means. Maybe not today that much anymore.
As of today it's GPU:
This distribution of workload today would mean GPU rendering. GPUs are supreme for some work. There are GPU clouds already in function. I doubt it you could do this with some Mac Studio hardware without some 3rd party dedicated software support. And I believe it's the same with yet to be announced ASi Mac Pro.
Apples CPU oriented xGrid has not been resurrected, is it? And with GPU capabilities in mind this time?
That would be a surprise. At least in light of those -comments I've earlier mentioned. "You can handle it like one processor only, no worries about complexity".
But the current M1's can't barely do a single x4 slot , let along plural x4 slots and miles aways from even a single x8 let along plural slots. 8K video capture cards are x8 PCI-e v3. two to three cameras that would be 2-3 x8 slots right there. Some folks have 'wider' x16, x8 PCI-e v2 or v3 cards because that is what it took back in the day before could go the "narrower" path Apple now prefers with PCI-e v4. Are 10-16k capture cards going to wait until PCI-e v5 to go mainstream versus go to x16 PCI v4 . Maybe , maybe not.
A 4x M.2 slot card is generally useful across a very wide set of workloads once get past the stage where just one (and only one) internal drive is sufficient. that was one of the other major issues they ran into with the MP 2013 ( and still have across the entire transitioned to M-series line up. Still stuck with the iPhone-equse just one internal drive is good enough mindset. Mac Studio is in the same boat. ).
The Afterbruner card swapped data over PCI-e v3 x16 data connection. Yes Apple has the codec covered internally, but ProRes (and H.265) are not the only codecs. Nor is Afterburner the only possible video decode card.
GPUs are debated louder in part to Nvidia versus AMD versus Intel arguments. But general utility across a wide set of workloads... they are not the 'only game in town' for Mac Pro oriented workloads.
With completely transparent software API ? Probably not for a broad set of applications. A 'RAM disk' exclusively for swap would only get engaged if Applications aggressively pushed past the max memory present on the system . ( generally apps should ask how much is out there and then allocate reasonable to the context large data structures and/or internal caches. ) Apple could probably more transparently assign the "RAM cache" to the file system 'read-only , write thru' cache structures.
Some 'RAM disk' kernel construct that could also possibly work with package RAM would be more credible. Specialized substantive kernel modification features that only work on a Mac Pro's hardware tangent SoC memory subsystem is likely a bit of a stretch.
macOS already currently will grab a substantive chunk of RAM and throw it at file system data caching now (along with memory compression ).. Apple could tweak that into a more complicated tiered hierarchy structure , but if they just wait a generation or two the package LPDDR RAM capacities would probably up and can just use a more mature version of what they got now.
WWDC 2022 would be time for some major memory changes though. So we'll see.
Not in the standard forms. But Apple isn't using LPDDR in the standard form. Over time Apple will leverage walking the increased density curve for RAM dies. Short term I wouldn't bet the farm on it ( die concurrency and package space compression with custom packages has been their focus so far. They really aren't shooting for maximum RAM die bill-of-materials cost. )
Assuming the ASi Mac Pro will come with dual 10Gb Ethernet as standard, a NVMe RAID card would be in the most need of bandwidth...
Audio cards (Pro Tools HDX) are good with x4 slots...?
One PCIe Gen4 (maybe Gen5...?) x16 slot & three x4 slots for the 2022 ASi Mac Pro...?
Samsung Develops Industry’s First LPDDR5X DRAM
Samsung’s LPDDR5X DRAM will provide over 1.3x faster processing speeds and consume nearly 20% less power than the previous LPDDR5 solution The LPDDR5X solution will broaden the use of high-performance, low-power memory beyond smartphones to AI and edge applicationsnews.samsung.com
As I have said numerous times in the last six months or so...
LPDDR5X DRAM:
- Due out late 2022
- 64GB chip density
Samsung Announces First LPDDR5X at 8.5Gbps
www.anandtech.com
That 64GB chip density doesn't seem to be "standard" .
From anantech article
"..Samsung’s implementation notes 16-gigabit dies (2GB) on a 14nm process node, with the company explaining that the new modules will use 20% less power than LPDDR5. It’s also possible to allow for 64GB memory modules of a single package, which would correspond to 32 dies. .."
The LPDDR5X only allows for banks of 16 dies. So have having 32 would be a specialized packaging of the dies of two channels each with 16 dies.
The 'problem' is Apple's "poor man's HBM with LPDDIR" implementation puts multiple channels into a single RAM package. Packing more current access dies into a single package will run hotter than just 1-2 channels and interleaved bank accesses. [ Eg. Apple will concurrently access die 1 , 8 , 16 , and 32 inside of a single package while the more mainstream LPDDR5 being discussed in these articles would only do die 1 and 9 in a very similar amount of space. The Pro, Max and Ultra RAM packages do run bigger than normal but there likely more die stacks in them too. So not dealing with crazy ultra tall (and expensive ) die stacks. ] So Apple may need to wait for 'reasonable priced' 32Gbit dies to move forward. Essentially Apple has already done the basics of the "get to 64GB trick with 16Gbit dies" to deliver what they are doing now.
No.
How much want to bet that WWDC 2022 has at least one "optimize for Ultra" session. This isn't really all that 'new'. How long did most Adobe apps not know how to leverage more than 2 and then more than 4 cores? Years. The M1 Max hasn't been out for 12 months. Not saying it has to take years, but that it really hasn't been that long since these 'big' Apple GPUs have been released. It appears have to deliberately throw a very large amount of work at these GPUs to have them fully 'wake up and get going'. Otherwise it is trying to save power on short 'blasts' of activity with 'good enough' performance.
It is indicative though of why Apple is likely not in any hurry at all to introduce alternative GPUs to optimize for on macOS on M-series. If there is a long list of Apple GPU tweaks and adjustments that needs to be rolled out, then they want developers 90% focused on those rather than have 'distractions'.
Probably not. Recent rumors point to the regular iPhone keeping the A15. On the iPhone Pro will get the A16. More than likely limited fab/wafer capacity is a contributing factor there. ( Apple can perhaps throw the full GPU variant at the regular iPhone that only last year's Pro got as some minimal upgrade differentiation. )
Pretty doubtful that Apple can both ramp on A16 and 3-4 of the M2 variants all at the same time. Mediatek, AMD, Intel , Nvidia are all ramping on N4-N5 variants at the end of 2022 also. TSMC is increasing N5 generation capacity but that is in the 25-30% range. It isn't being doubled.
An M2 Ultra/Extreme would be surprising as opposed to waiting for M3. If the Mac Pro isn't going to ship for over 6 months then it doesn't have to be an M2. (nor does M2 have to cover the whole range. Even more so if M2 is just TSMC N5P. ).
Apple could do some M2's and a few hyper expensive Ultras . An extreme pricing on the Ultra's would push their volume down. And if there is no M2 Max 'singles' to sell can keep the MBP's rolling forward with M1 versions through end of 2022.
The hype train that yearly Mac updates are gong to come to Macs ... nothing so far points to that. And there are product chassis changes coming also (Air , Mac Pro , Mini etc) ... Apple doesn't tend to walk and chew gum at the same time very well where multiple concurrent Mac product changes.
( story today that Apple is telling suppliers to chop iPhone SE parts as sales are slower. High inflation, wonky supply chains , .... This post second coming of Jobs version of Apple has never navigated something like that before. It is wrong time to throw crazy product variations out there hoping folks just keep throwing money at Apple with no regard. )
There is probably enough if wanted to do something different. Just not enough to make it all 100% uniform, unified memory.
For example if took the total 2.5TB aggregate of UltraFusion and split it 4 ways. That would be 625GB/s for three other dies and a PCI-e block. Even if just took 2TB (some overhead creep) and split for ways it would be 500GB. Could that look like just one GPU .. probably not. ( down under NVLink 4 range).
I highly doubt there is just one 2TB ring/mesh bus running between the chips with 10K pins/pads. Pretty good change that is a number of busses that could be pointed in different directions for a more non uniform system. The problem is the memory and interrupt subsystem won't scale. They'd need a new implementation.
There is enough bandwidth between the two CPUs that if put some PCI-e provisioning block" inbetween they could get out of the crippled PCI-e lane provision corner they have painted themselves into to do a couple of slots for a Mac Pro.
The M1 Ultra is relatively lame for a small tower system with slots. (unless they have once again hidden a significant amount of silicon from view in the talks about what they are delivering. )
By TMSC N3 it seems more likely Apple would collapse an Ultra into just one bigger chip (but still smalller than 3090 ) because that will be better Perf/Watt than using UltraFusion. At that point can just join two of those (so still only using all the bus bundles to do one CPU to CPU 'bus' ). After that just ride fab density improvements into more affordable big, then later upper middle size, dies in a pair.
It could. If the M2 is just TSMC N5P process (like the A15) there is not much of a big blocker in doing a large die solution first and then mid size (or smaller ) versions later. It is a relatively very mature fab process so can make big affordably. The bigger issues would be whether they were running into wafer start competition with the A15 or A16 (if on TSMC process that is matched to ).
Similar issue If the Mac Pro waits for M3 (and TSMC N3? ) process. Could use a $6-8K package as a pipe cleaner for the process node ( lower volumes but far higher package prices to pay for all the 'defect' dies. )
Always smallest to biggest doesn't really make sense when the Fab processes iterate at a 18-22 month rate (Moore's law with some slow down). Sometimes the big dies can go first and pipe clean the process and some time the little dies can go first and do the same thing. The A-series and 'plain' Mx don't always have to play the bleeding edge process role. The yearly schedule of the A-sereis means they can't match that 18+ month cycle at all over time on each year.
If M2 is just TSMC N5P though it doesn't make sense to me for the very large dies at all though when N3 isn't that far of a wait. Apple could either make the Ultra just one incrementally larger die (at better Perf/Watt than using UltraFuion) or an easier package with closer to mid-sized dies.
If M2 is a TSMC N4 variant like the A16 then initially skipping the Pro/Max and doing a M2 and a relatively low volume M2 Ultra could make sense. The N4 is really a N5 tweak with some incremental density improvements for a limited set of things.
The plain M , Pro/Max , and Ultra are in different Mac Products that likely come out at different rates over time. ( Mac Pro 2010 [fake 2012 ] -> 2013 -> 2019 is no where near even 2 year cycles. That is probably not going to change with transition to Apple Silicon. That was basically primarily Apple effort/priority allocation all along. Not Intel's. )
Seem "Pre enough" for Apple. At the Studio introduction there were comments as to 128GB was a "insane" , "extreme" amount of memory. Now, they are trying to skew the context to "insane" for a GPU direct access. But the SoC is basically a GPU with some other stuff wrapped around it. The core infrastructure there is primarily for the GPU , NPU , ProRes; not the CPU cores.
In 2023 or 2024 they'll probably take any 'easy' 256GB jump, but I doubt they are in a hurry to go through gyrations (and higher Perf/Watt) to get there.
For folks with 500GB data sets even 256GB isn't "pro enough". I suspect Apple has a pretty good idea of how many folks they are 'cutting off' with a 128GB limit. Given the iMac Pro went up to 256GB it is a bit low even without shifting to the modern upper Mac Pro user base.
Apple's bigger problem solution would be getting to 128GB more affordably than cranking out a 256GB prices that will just drive away even more classic Mac Pro userbase. The non modular memory is going to piss off a number of those folks and the 'Apple tax' prices will do even more. Apple's hope is that get into a 100-200GB of HBM2/3 versus Apple "poor man's HBM" kinds of value comparisons.
I'd doubt there will be M2's before Fall at this point. The upper Mini and Mac Pro are still on Intel. At least one of those is probably next off the board (probably the mini) . The overall breadth of the M-series line up probably isn't going to be at the same pace. Or even touch each generation.
The last several A__X iterations just went on the evens when has a full fab process node upgrade available. (enough a density increase to add some hefty upgrades ... not relatively small tweaks around the edges of performance).
M2 generation could be just M2 and M2 Pro. Then roll the M3 sequence out in 'reverse' order. Extreme , Ultra , Max/Pro , 'plain' M3. Go to another limited subset on M4 and they a broader range on M5 generation.
I honestly don't see the problem with Apple continuing a partnership with Intel and AMD for the Mac Pro. Apple Silicon is not really workstation material. I think there are a lot of customers that wouldn't buy it. Like it or not, the x86 environment is the gold standard for business. Take a trip to your local business school. I doubt you'll find very many profs that use Macs.
Mac software economics. Both Apple and third party developers are going to have issues porting to such a minuscule small user base over the long term. Detached from the rest of the Mac line up, this is user base that probably less than 200K or even 100K. In part , same reason don't see the same breath of mac developers doing apps for desktop Linux. Also related to the 'chicken / egg' quandary that Windows on Arm has growing software catalog.
The x86_64 Mac user base is around 100M now. 5-6 years down the road that will be far closer to just a 10M (or less). And in terms of active 'new' software buyers much closer to 1M or less. (Mostly going ot be a market of folks who already own something and just want a machine to keep running in the same 'time bubble' . Another segment of time, the user base will drop lower than that. The user base will implode. The Mac Pro isn't going to reverse or stabilize the trend.
Most new software startups aren't going to target imploding user bases. Where there is subscription software revenues there probably won't be a wholesale stampede out of doing updates. But the base of folks paying the subscriptions for the x86_64 is going to also shrink. For the non subscription group mostly have a bunch of folks who want someone else to pay for their upgrades. If vast majority of those "someone else" folks are paying for macOS on M-series upgrades then ... money talks.
If Apple was a different company and there were two or three other x86_64 Macs kept in the line up that had substantially more volume in sales then maybe. However, the Mac Pro by itself isn't viable. In the 2020-2022 time period it made some sense to do one last refresh while the x86_64 userbase was at max size. At this point, the user base shrink rate is pretty close to full speed.
'a lot of customers' relative to the overall Mac market won't buy? No. Not even close.
'a lot of customers ' who bought an iMac 27", iMac Pro or Mac Pro 2013 won't buy ? Nope. The Mac Studio is a very competitive upgrade there.
'a lot of customers' whose budget only went up to around $3-4K who used to buy Mac Pro's back in the 2009-10 era? Largely No. Any upper 10% Intel/AMD Mac Pro that would be left over coverage that the M-series doesn't cover. The entry point would likely be above $6,000 (if not $9,000 ). Apple has already said they are going to transition the "Mac Pro" to M-series/AppleSilicon. It probably won't have all the modularity features of the current Mac Pro 2019, but it probably will at least subsume the $5-7K price point.
[ there is the external xMac position that Apple should have some $1,500-3,000 slot box that has generic PC market modularity. That is basically doomed because Apple has explicitly said they will replace what they can cover in performance. The low to mid range of Intel/AMD range Apple can cover CPU wise. ]
At the Studio introduction Apple stated the most popular Mac Pro CPU was the 16 core and the most popular GPU was the W5700. Narrowing down to just the Mac Pro space, there are a large number of folks who are > 16 cores and/or > 57000. But many are not. Once the Mac Pro user base are stripped of the folks for which as Studio is good enough. A 1-4 slot , 'half size' Mac Pro strips off the folks who just need some PCI-e card modularity (storage , network , a/v input/output, etc. ) what is going to be left. Saying there are 45,000 left out of a former pool of 70,000 is not necessarily a viable remainder. If the pool starts off at the edge of viability then any substantive cut to the pool can push it into the unviable zone.
The main product of a business school profession is journal articles , case studies , and lectures. They deal with a barrage of emails and research/student communication systems. If cranking out journal articles and lectures all day long then Macs do have a substantive user base share.