You'd hope that the last MacOS for Intel has no time limitation on it and will continue working for many years.
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 What does Apple's switch to ARM based processors mean for the future of the Mac Pro
- Thread starter Korican100
- Start date
- Sort by reaction score
Wow, thanks guys. Exactly what I did not want to hear. And new machines utilizing only Apple GPUs - good god!
Well they should be pretty good at rendering emoticons at the very least.
Well they should be pretty good at rendering emoticons at the very least.
Just going through the WWDC videos. Apparently, if you’re not using DriverKit, you should be.If you’re not, pain is assured. Another interesting thing that I keep seeing over and over again in the videos I’ve watched...
View attachment 927578
Apple Silicon Macs are ONLY defined as utilizing an Apple GPU. Only Intel-based Macs are said to use Intel, NVIDIA or AMD GPU’s (btw... NVIDIA??). What this says to me is there won’t be any non-Apple GPU options in Silicon Macs.
iGPU only Macs : MBA , MBP 13 , Mini ( Macbook if bring something in that space back) , entry iMac --> 3-4
discrete GPU Macs ; iMac (rest of them) , MBP 16" , Mac Pro ( iMac Pro if count seperate) --> 3-4
If in first 6-10 months only aim at the first category then .... that chart would make sense. It is a huge leap though to label it a "for next 2-4 years" chart. They'll be another WWDC next year and they can do same presentation with new info. ( something to do
) . Thunderbolt should open the door to an eGPU . If it is a DTK era chart then makes even more sense. There is only an iGPU and can't possible get another one into the picture with only USB-C 3 data ports. (at least not without a decidedly non mac kludge. ). The chart is probalby dated but it is accurate for most developers for the next 6 months (at least ).
I am inclined to think though that they'll include their GPU in every SoC that they do. Perhaps not the "biggest one" in the SoC's that are enabled to handle discrete GPUs but something so they can run their nominal system code ( graphic accelerated login screen) against with the early boot drivers and get standard image and video/picture fix function logic. (similar to the T2 graphics + video + image subsystem being uniform on those systems now).
Just covering the full range of Mac CPUs is a questionable reach for Apple ( without shrinking the range of Macs... e.g. dropping upper end Mac Pro space). Trying to cover all possible GPUs also would be lots of Cupertino kool-aid drinking. They probably don't have that kind of resources assigned, nor would it work economically at all.
I’m just posting what’s in the videos.
Would Apple provide outdated information in a presentation meant to help developers get up to speed on the new architecture? Maybe? Great thing is, anyone can watch these videos. Every video I’ve seen about Metal and the graphics on the Apple Silicon Mac mentions the GPU being on-die with the CPU (and the benefits thereof) and they’re instructing developers to code to the Apple Silicon Mac assuming that’s the only GPU option, just like on iOS. There’s a couple videos they’re not posting until tomorrow, those will likely provide additional info.
that notion suffers from the deep seating problem in that the A12X did not ship at the same time as the A12Z. For a vast amount of the production lifetime of the A12X there was only one "bucket" to drop the finished dies into. (other than into the trash can bin. ). The notion of sub selection into multiple product categories has two major problems.
First and foremost, there has to be more than on product category to separate into.
Second the defect rates are usually way lower than the production of parts needed for a "subclass" bin. Typically 30-90% of stuff 'binned' into a lower product category works just fine. The 'binning' on a process node that is 87+% good rate is mainly just goosing 1-2% more product out the door. Decent for making some extra "nickles" but pretty bad as a reliable source for product that is 10-30% of the dies produced.
For the first round of iPad Pro based on A12X Apple was mainly just goosing the margins a bit on the product by dropping at a GPU core. More of them were good and a narrow few bad (but there use taps the margins up a bit). The iPad Pro was so far out in front of other tablet performance they could afford to roll out with slightly kneecapped host processors because nobody else was going to outperform them.
That also gives them a "free" upgrade later with the A12Z ... Hey look! new iPad better get this one becaue 'Z' is bigger than 'X'. .... and it is almost exactly the same SoC component costs for Apple. ( if 14X was going to slide then pulling this "little Joker" trump card is all the more effective. ).
If a new AppleTV ships with a A12X currently with the A12Z then might get into some binning for slightly goosed margins for the AppleTV chips. There would be two concurrent buckets of volume production.
There is a decent chance one of the Mac Products might share a die design with the next iPad Pro. ( e.g. have a x4 PCI-e controller that don't use on iPad Pro but do use on the Mac.). Binning is probably going to be a bad way to view that though. Likely largely some subcomponents will be "fused off" that aren't broken at all.
If Apple is putting substantive I/O abilities on the Mac version though the "fusing off" of function units will be on a separate die from the iPad Pro.
I’m just posting what’s in the videos.
For the DTK , which is all they have access to, the information is not outdated in the slightest. Given the DTK is only system shipping for probably at least 4, if not more than 6 , months . There is no problems generated for that code development for that time period. If the first Mac out of the gate is still Thunderbolt less then the chart will still be highly accurate for any deployed applications on those systems.
Some much narrower set of developers who sign a strict NDA could easily betting a better chart. Fooling yourself if think all the public released material hasn't be combed from classified information.
the whole point of having the DTK is test actual code. If there is zero way to test calling a AMD GPU from the DTK what is the developer going to accomplish with the system. Nothing. It is a waste of time to construct a code branch that doesn't exist. Spending that time making the Intel side work better with the others will probably pay off better.
Honestly, the real. very, odd duck there is more so that Nvidia is on the list at all. Not that the "Apple GPU" is in the other column. If want to engage in cryptic reading of vague tea leafs that might be more productive. Apple being 'nice' and leaving the door slightly open. Or legacy historical chart artifact. Or ....
Probably more so mentions that it is in the package and "unified memory" more so than blow-by-blow die characteristics. But yeah. much there points to Apple probably going to attempt to make the Apple Silicon cover as many iGPU instantiations as they can in the Mac line up. They are going to have to give on that a bit at the upper end of the line up.
Again. Right now that is all they have. And all they are going to have for many months. If Apple is super eager for people to pound 100's of man hours on arm64 code over the next 2-3 months there aren't any other options. I don't think the objective of the WWDC sessions are for folks to sit around studying there navel for 4-5 months before they start working on this. By december Apple would like to have a sizable fractions of the more popular apps done.
Getting developers to move is like herding cats. If sit there as say you can do any complete testing for 4-5 months then probably not going to get the bulk of the code done before then.
And frankly if spending $500 to lease a computer for 5-8 months... before have to give it back. the developer is wasting money if don't put in the time.
Maybe my ideas got lost in translation. I will try again.I think some of that though is that Intel were going from strength to strength at the time and compared to their earlier spec'd chips of that generation, the Intel seemed to be knocking it out of the park. I still have a bunch of 2011 era gear that is hitting the point where I want to replace in the next year so I'll personally be holding out for ARM because I hold to the rule that if you're going to buy Apple gear, you buy it when it's just released.
Isn't that time stamp the end of the video? Feels like a troll
Whilst it might look like I'm disagreeing with hardw0od, I echo the Intel transition experience where PPC fell away rather quickly. Apple in the same situation dropped the legacy architecture quickly and really as a programmer I don't blame them because it simplifies development and testing, plus for low level systems you get to clear out the architecture specific code. I actually think it is likely that the Intel Mac's will get two releases after the last one shifts over, likely the Mac Pro, and that'll be it similar to PPC. For me the unknown is if they'll actually ship a replacement ARM chip for the Mac Pro in the two year time frame they're talking about. It does mean if you've bought a Mac new in the last year, you'll likely get five years out of it before you get no updates.
The other reason to drop support is to add new features and simplify existing implementations. We're already starting to see out of WWDC notes on how low level security will look different for MacOS with ARM for example. They'll be able to engineer into the silicon functionality that won't be there on the Intel chips. That I think is the greater push to obsolete Intel support more than anything. From that same keynote, it's about products.
There were some posts stating that the 7,1 would be "obsolete" in two years.
In my dictionary the meaning Obsolete => No longer in production
I don't believe that the 7,1 will be stopped in production in two years.
Also, Tim Cook indicated that there are more intel macs coming soon. This was stated at the end of the video link I gave. Just back up a minute or two from the end of the video it you are having some nuance. This means to me that Apple will continue to sell and support intel macs for (its the meaning) many years to come (I suggested between 7 to 8 years for the 7,1)
This is the problem that has puzzled us all. They did not speak for many years. They said about the just years to come, and this can be understood in different ways: 2-3? May be 5? Even if 5, it is not very much.
What we are seeing here is the end of Moore's Law in the 2010's, so the performance of desktop computers from the early part of the last decade are still reasonably good when compared to the latest models. This wasn't so much the case in the 2000's.
The same thing can be seen in data centres, where there are racks full of Sandy Bridge Xeon powered machines, because they are still good enough for the job.
You are somewhat fortunate with your Mac Pro 5,1 as I think you can run a recent version of macOS if you have a Metal-capable GPU. Others can't upgrade beyond macOS High Seirra, despite Ram upgrades, SSD upgrades and having a decent enough intel cpu.
Staying with an intel cpu isn't going to protect our machines from obsolescence.
Apple said both in the keynote and in the press release that the transition would be complete in about 2 years. Technically two years from June 22 2020 ( June 22 2022 ) , yeah I suspect there is a probably a good chance whatever Mac Pro is current at that point will still be in production. By end of December 2022, then no . That isn't what Apple is saying. Like the MP 2013 , iMac Pro (2017) , and MP ( 2019) , we may get a "sneak peak" in June 2022 and launch in December 2022 , but 7,1 probably will be ended somewhere in June-Dec timeframe. Moving the "last" Mac Pro to the other side of the 2022 Fall Mac OS release will help.
Apple is only giving themselves wiggle room over it is exactly 12 months. They may need some extra months for slop in the schedule. ( They certainly have in the past and probably will still need that slop two years from now. ). If everything happens to go perfect and the Silicon , macOS , 3rd party GPU vendor , , etc etc. all finish early and there no more major pandemic disruptions , then maybe they get done in less than 24 months. Especially with the pandemic looming it would be looney toons to put themselves on a super rigid hard deadline into a somewhat uncertain future.
Support for many years to come? Yes. Sell? "many" is not a good connotation. ( other than as refurbs . )
There are more coming but they also are going to stop coming in "about two years". There is a significant problem though in what is left at the end. If it is just the Mac Pro that probably isn't going to help a whole lot in extending the support. Apple can't walk away from a large bulk of 86_64 mac customers. 100M 86_64 Mac users right now. There will probably be more than 100M in 2021. That number probably won't start to even substantively shrink until well into 2022. But if Mac Pro sales are less than 1% of sales they'll be doing practically nothing to slow down that overall shift in the userbase. Once the top 4 high volume products get flipped there is extremely little the slower moving Macs can do to extend out the shift. If the Mac Pro is last, it will be like trying to use a single sandbag to hold back a Tsunami rolling onto shore. If Apple gets all of the sub $1,700 laptops and the Mac Mini all switched over by March-April 2021 that Tsunami is going to be pretty tall by end of 2022.
Pretty good chance that Macs needing discrete GPUs are going to stick with Intel for next 12 months or so. Apple's first stab at a Mac arm64 will lean heavily on the Apple integrated GPU. Apple could prune off a subset of the iMac but for most part will need Intel for that. iMac Pro is overdue. MBP 16" could get an update based on Willow Cove depending upon how it goes.
If Apple squats on the Mac Pro for 3 years though then the 8 years from now I think is a bit of stretch. Especially, if they come out with something reasonably competitive with other workstations 2 years into the future. This Mac Pro isn't going to look so hot ( in terms of performance. it may appeal to people who already have them on "sunk cost" metrics. ). And two years after the transition, it will be worse. The number of folks who will 'bail' after after 5 years on the system is probably going to be a pretty high percentage and the base percentage of overall Mac market starting with is in the extremely low single digits ( if a single digit). In 2-3 years after the transition probably going to start seeing apps that lean heavily on Apple tensor cores for ML and some other features.
8 years from now, only if talking support of the "high security bug fix patches" extended back to older macOS releases. It isn't going to quickly fall off a cliff, but extending out the Mac Pro for 3 years until upgrade is going to be a bit of a dual edged sword.
No. This isn't that vague. The years aren't where folks are confusing themselves. It is support versus selling product. Apple said transition would be done in about two years. that part has to do with selling. If the transition is done then there aren't any more Intel ( x86_64) systems.
How long support them after the transition is done is where Apple is using the "years to come". So after 2022, how long. Some folks are speculating 3 years. ( Apple kneecaps the Intel systems after 3 years with no new upgrades ( still will have about 2 years of security upgrades though) ) Some folks are saying something closer to the normal Vintage/Obsolete countdown clock ( 5-7) years. ( and again about 1-2 years of security fixes ).
The estimates based on the PPC->X86 transition are probably not all that good for two major reasons. First, is that Apple made that transition in much less time. ( basically done in about 12 months versus probably over 24 months this time). Second, the installed PPC Mac base was an order of magnitude smaller than the current x86_64. There are 10x as many active users Apple has to move over this time. That isn't going to go faster either. Most folks are using their classic PC systems longer is a third major factor ( people sinking more money into cycling cell phones / iPads . and generally PCs being fast enough for vast majority of folks for mainstream uses . If the Apple Silcon Macs are alot better they might get most folks to go back to those older update rates, but nominally those rates are in a hole relative to 2005-6 ) .
If Apple sees a large stampede into Apple Silicon Macs then they may shift to a "shorter" window ( like 3-4 years). If they don't then cutting off x86_64 mac users makes even less sense than it did in 2005 since Apple is getting a Services revenue from those folks even if they don't move to Apple Silicon. Pissing off significantly large numbers of folks (who cancel subscriptions ) would damage both the Mac and Services business. Walking away from re-occuring , steady stream revenue is plain goofy. Apple probably won't "save' that much money dropping down to one build until the number of folks still on x86_64 is small enough. And that is just going to take more time this transition.
That is is where the puck has already gone in vast majority of "embarrasingly parallel" workloads. If go the the TOP500 supercomputer list and yank the GPU cards from all of those systems, the large majority of those systems drop out of the TOP 500 list.
don't need to get to Dual GPUs to get to one or two order of magnitude more cores in a GPU than CPU. Even more so for graphics oriented problems where the result is going to displayed anyway when done with the computation. There have been hundreds of posts in this Mac Pro forum about how Apple is doomed if they don't have CUDA . that isn't just Apple pointing at GPUs for a large share of the compute that is customers. To imply this is some future thing that nobody is doing right now was dubious back in 2013 when Apple rolled out the dual GPU system. At this point though this point it is just disconnected from relatively. Heavy parallel compute has moved over to GPUs in a extremely significant way. Nothing about " guessing where puck is going" at all.
If Apple doesn't provision the ability for multiple discrete GPUs in the next Mac Pro then it will fail horribly. Period.
So Apple pointing at discrete GPUs for "embarrassingly parallel" computation work will be the same thing most of the customers are doing also.
In terms of performance per watt efficiency GPUs generally beat CPU cores. Those cores are usually smaller and more focused just computing rather than branching and non compute tasks.
If it is Metal vs CUDA then Apple has more than enough money and patience to push that boulder up the hill.
The not so encouraging sign is that Apple hasn't turned the corner on taking OpenCL off the deprecated list. If OpenCL stars getting more adoption traction again and Apple bails they'll have problems.
But isn't just Metal. If Apple wraps the acceleration libraries around some chosen 3rd party GPUs ( AMD and Intel) then . Alot of the 'noise" about CUDA is about libraries of code, not super custom code. Metal isn't behind as much as the libraries are behind.
You are looking for ham from a strict orthodox Jewish kosher store. You can have an opinion that they should serve ham but they don't believe in it. It isn't mainly about the product.
Apple isn't going to talk extensively about a product is probably two years out. ( Or maybe one year out and the "end of the line" for Mac Pro over the long term. if sales are so bad they don't really want to do it very long term anymore. Release a 8,1 with Sapphire ridge next year and except from the "transition" because end of the line. I don't think that is a likely path though. I think Mac Pro is coming at the end with whatever they come up with and they'll just "sleep" on the Mac Pro for 3 years. That's fits the pattern on the iMac Pro and MP 2013. )
If stand on your head and look at it funny perhaps. but Apple handing over the old "design mule" boxes they have been using for the past 1 (probably closer to 2 ) years to outside developers is pretty indicative that they have other , even newer boxes for their own folks to use now. I
So the Macs coming out will somewhat not be version 1.0 products. Perhaps something akin to v1.2 or v1.5 . Apple has been using these for a long time that means they have made progress. That not getting some half baked macOS sholved out the door too fast just to support a 'dog and pony show' pizzaz.
The DTK is for the masses. Vendors who have had super tight NDA beta boxes before will probably get access to not widely released hardware in the Fall sometime.
I doubt the Mac Pro version Apple Silicon is even out of finished design phase this year. (tape out probably some time late in 2021 ... maybe. ) There is nothing to hand to folks even inside of Apple let alone to 3rd parties. I would expect that the 27" iMac stuff that has at least one discrete GPU (either directly or through Thunderbolt ) to get some more "preview" talk time at WWDC 2021.
[qjuote]
Where their GPU work lands is important compared to Intel iGPUs. If they can make Intel look bad. Great. [/quote]
It is going to be harder to make Intel look bad by the end of the year. Intel is about to make a pretty big jump.
https://www.tomshardware.com/features/intel-gen11-core_i7_1065g7-tested
The A12Z isn't Mac Apple Silcon. In one of the WWDC sessions recently one of the presenters mentions that the DTK doesn't exactly run macOS fully correctly ( that is a compatibility note in the "release notes" distributed with kit.
Getting super twisted about A12Z limitations is an overreach. No Macs are going to ship to customers with the A12Z.
Proably no Macs will ship with something labeled A14x either. If there are no discrete GPUs coming for 9-12 months then what are they talking about that is relevant to developers now for the next 2-5 months?
but that isn't likely to be the first arm64 Mac out the door. I suspect the Mini is much higher in the priority queue than the 27" iMac. Apple may tackle the smaller screen iMac with their own. For the entry level , "edu" iMac they can certainly cover that in scope. Maybe the bottom of the Retina series too.
The mini's would make sense because lots of folks do build-test farms out of Minis. If there is tons more arm64 code to test then going to need lots of build-test farms for that code. That those build-test farms can do a ton of native iOS-iPadOS testing too is even more "extra" gravy on top. Apple would sell lots of those very quickly even if just brought the DTK up to "full Mac silicon". That would be the best desktop compliment to release with a moderately high laptop . The dGPU-less laptop is an even better fit.
dGPU sysetms is not what they need to get first out the door at all. Once in the dGPU space Intel's solutions have far more traction in competition.
Demonstrate how? Look if they already did this with the new 2019 Mac Pro in terms of saying "on board with the form factor". If they are doing something like that then dGPUs are a given. It coming. I don't see have there could be any reasonable disconnect there. There was no qualifying statement that they were only going to do a smaller subset of Macs in two year and would be dropping some. They basically said they'd transition the line up in two year. The Mac Pro is in the line up. So Mac Pro should be in the transition. There is not much to point to to say that it not given their baseline disclosure rules.
The bigger die is more complicated and will take longer to build out. That also means it is more probably not done. Which if it isn't largely almost completely done , also means Apple isn't going to talk about it in detail. that worked so well for them to violate that to Air Power. Not sure why they'd break the rules again now.
No they are not. The A12Z is almost two years old. These DTK are probably a two year old design also. Add that 24 months to the 24-28 more months and you get a 4 year window. Getting a new CPU design out the door in 4 year is pretty regular. That is doable is assigned the manpower , simulators, and equipment to the problem.
If they use basically the same central core unit design for the iPhone as this larger chip then that lightens the resource loads. Need to tweak memory controller, PCI-e controller , USB, SATA , classic I/O handling , etc. but that is running concurrently with the iPhone cores than isn't a huge stretch goal. ( if pulling Thunderbolt4/USB4 inside that couldn't have started "early", but that can be punted to discrete controllers relatively easily. ).
I think is is a bigger jump not because of the technical problems entailed but how Apple tends to under resource top end Mac product development. That much smaller sales volumes probably tend to drive lower resource allocation.
I don't think it is a matter of an AirPower thing where it won't work at all. But they may take another step back from trying to go "head to head" with the higher end Dell/HP/Lenovo/Boxx workstations.
If Apple is too constrained on resources they could attach a small Apple Silicon SoC to the board where the T2 goes (and put it into a T-series mode ) and get an off-the-shelf Neoverse solution to toss into the Mac Pro. They don't "have to" do the bulk of large core count CPU design if they don't want to. That are ways of "mailing it in" that largely can just buy. Apple used to slap "G3" , "G4" , "G5" on chips that weren't called that in any other systems. They can get an Apple slapped on someone elese stuff if pay for it.
Given the stuff seen at WWDC I don't evpect it likely but making the date shouldn't be hard is they work at making the date if just lean on other folks stuff for bigger parts of the ARM core subsystem part.
But they didn't had out a working prototype at the "pow wow" meeting in 2017. That is a significant difference.
It is not the lag talking about here at all. The gap between when had that early prodiuct range covering prototype ( probably on order of two years ago with a A12X sitting in same spot as A12Z doing not that much less) and spending another two years getting the Mac Pro silicon deeper into the design cycle before making the call on timing is indicative that it isn't the same at all.
Something for the mid-range Mac probably are at the tape-out stage (or very close). Once there and talking about wrapping up on the biggest one over next year or so then should know reasonaby well on track or not. Thrown in a 6 month "fudge factor" so ending before 2022 is over ( still in "about" two years zone) and should e doable with a highly dependable group. Even more so if farming out bigger chunks of the I/O subsystem to folks who already do that and just coupling via PCI-e ( discrete ASMedia and/or Intel chipsets for USB/SATA or Thunderbolt4+ )
Apple inscribes "design by Apple in CA " (or Cupertino) on their motherboards. I was deeply puzzled back in 2017 when folks speculated that it would on be a couple of months to a new Mac Pro because they were just going to throw something into the old case and run down to Fry's to find SATA to slap in the machine. I thought it would be about 18 months. It is was bit longer but yeah.... they were probably going to dllberately design something.
Why would Apple not deliberately do something. It is is just a matter of getting making something for everybody and whipping lots of different products out the door then sure buy lots of stuff off the shelf and shoe horn in the compromises the save time and money. Apple wasn't trying to "save money" on the Mac Pro. The entry level price went up around 100%. They were going to be leaving more than a few folks behind.
pretty decent chance that this Mac Pro baseline has some broad thoughts weaved into it for the Apple Silicon. Those would have started around the same time. No dual CPU packages. Some tiny SATA could add with a small discrete chip if necessary. Extend the host USB provision out through the I/O card. Discrete TB controllers. Some one the GPU cards ( DisplayPort 2.0 will be on future cards ) The DisplayPort output from the GPUs cards flows into switched internal network ( so won't be too hard to add an Apple iGPU into a slightly bigger switch with a couple more inputs ).
The Mac Pro doesn't have any more "regular" USB sockets than a Mini or iMac . If have the basic ports support in the SoC for the iMac then can have basically what need for the Mac Pro. Mac pro probably needs some more PCI-e lanes out to fill in some "extra gaps' with perhaps a few more discrete I/O augments but have more space to put discrete components to "augment" that SoC.
There is work to do with more re-drivers because the many of the I/O channels on the larger board will need help at higher bandwidth settings ( PCI-e v3 versus probably v4 on first iteration. ).
We still have to guess how Apple will behave in this situation in practice. But your words sound reasonable, thank you.
Pulling Thunderbolt inside the package without having a GPU easily couple it to is an oxymoron. One of the principle points of pulling it in it simply and make more efficient and easier for the system vendor to do ( that latter because it is almost basically done. )
On a Mac Pro System with a non gimped 580X card at least half of all the thunderbolt ports on the system are on the GPU cards. Routing the 580X all the way back to the CPU package and then all the way back out the outside ports on top and I/O simplifies things how?
The laptop targeted dies yes. The larger boards of the desktops that doesn't really make much sense at all. And implementing Thunderbolt takes up a significant amount of space. If trying to crank up arm core count there is a inherent floor plan space usage conflict there. Apple probably needs at least three different SoC basic plans . iMac 27 , iMac Pro , Mac Pro doesn't really line up with that.
Apple does have "accelerated graphics" for the login. It doesn't make much sense to have a wide side of EFI boot stage graphics. Something substantive for Apple and then a basic dynamically loaded one that can just copy the frame buffer to a discrete card of the finished accelerated graphics. That more more simpler (work wise) and more secure.
Similarity if someone pulls all the MPX modules out, then if there is iGPU then the Thunderbolt ports still work as Thunderbolt should. If trying to provision Thunderbolt uniformily across all macs then embedded GPUs all the time would make that uniform across the board.
Apple could take One big and two four smaller chiplet approach to put the PCI-e v4 fan out ( and memory controllers ?) onto those chips and leave the Big (P) , little (E) , Neural , and GPU cores still coupled around a same die system cache. No NUMA introduced into the cache coherence. There are different ways on how to group what gets kicked out and what stays in the base "hub" block in the middle.
What AMD is doing is better way to scale to 64+ cores. Apple probably is not going there. macOS 11 probably isn't going there. So the SoC probably won't either. I'm not sure Apple's system will work well cache coherncy whise when the Nerual cores are seprate latency than the arm cores. Or different E cores are on different latency than P cores. And it really , really , really doesn't make much sense to scale up E cores at the same rate as P cores. If each one of the "cores only" chiplets had both then ...... what single user workstation workloads really needs that many E cores?
The I/O block is what is varying across the full Mac range as much as what is changing with the cores. The memory capacity is even bigger variation. From double , triple , quad digit capacities . Not going to have a single common core for memory controllers.
When AMD goes to APU zone is it all one monolithic die. ( because that saves power. ). Saving power is a firm underlying baseline objective for Apple Silicon. I suspect there will be some multiple chip module stuff but not necessarily to create the biggest number of possible CPU count differentiated packages. That isn't the stated primary goal.
For laptops. Err no. AMD isn't even doing for laptops. Same basic reasons.
I think they are making new dies with ARM cores on them. There is almost nothing to indicate that they are making "ground up" new arm cores for these. Perhaps something tweaked to a higher power thresholds but the same implementation is probably present everywhere. Remember one objective is the run iPhone binaries on the Macs. So same stuff that iPhone binaries expect out of the arm cores should still be there on top of a very similar OS kernel and libraries.
The iPhone/iPad Pro cores are likely the primary drivers of the architecture and the Mac cores are tweaked derivatives on that. Mac ones will have more stuff coupled into the package but where Apple can get away with the same internal bus fabric they probably will. They made need something of a upgrade at the iMac -> Mac Pro level but again likely an iterative derivative.
If they go with this approach it will result in lower prices for them and extremely high yield rates of all the dies they have manufactured
Intel's 10nm+ server line was suppose to transition to EMIB and smaller dies coupled to together. (that was the projection back around 2017
"..Update 1: On speaking with Diane Bryant, the 'data center gets new nodes first' is going to be achieved by using multiple small dies on a single package. But rather than use a multi-chip package as in previous multi-core products, Intel will be using EMIB as demonstrated at ISSCC: an MCP/2.5D interposer-like design with an Embedded Multi-Die Interconnect Bridge (EMIB). .."
Intel Confirms 8th Gen Core on 14nm, Data Center First to New Nodes
www.anandtech.com
Along the way Intel shifted to trying to stick with the higher clocks and that didn't work out.
Everyone is trying to make increasingly denser dies. Intel's probably was how far trying to leap in one hop. That was a 'bridge too far'. But there weren't on a bigger forver path for substantially long amount of time.
The issue for the Mac Pro Apple Silicon is that is would be coming at either a mature 5nm or a 3nm jump in point. The denser is already there and if only shooting for a 32 core cap that won't be "max reticle" sized die. Especially if prune off some of the I/O and spin out some of the very large System cache ( or add some L4 to relatively moderate cache to make it "feel" bigger to the cores drawing data from it. ) . Large cache RAM soak up gobs of floor plan space.
No previous macOS version has had that kind of loose constraint. Even Windows really pragmatically doesn't. And almost nothing Enterprise that doesn't have an eye-popping annual price tag attached to it.
macOS instances are going to expire. That rate isn't really a deep mystery. ( about 3 iterations after superseded for last of a subset of the security updates. )
that hardware sold earlier in 2020 in isn't going much past 7 years after the last sold date. So fixed time limit on that.
There are limits on all of them. has been for decent so not sure why someone would try to set expectations otherwise. Why Apple would do lots of work for free ( no pay) forever .... not going to happen.
If the support last 5-7 years after the transition. ( relatively easy if loop in the extended security updates and squatting on a 'last' macOS version for 1-2 years. ) then can get even more by just firewalling/disconnecting from general Internet and keep going for another couple of years.
It is not just Apple that is unilaterally setting the dates. If Mac Pro sales fall off a cliff and go no where for 2 years then Apple will probably put less effort into supporting it longer past the transition. If Mac Pro sales stay as constant into 2021 as there where in 2020 then there is less of a chance Apple will cut them off "early".
If the Apple Silicon Macs come out and sell at about the same rate the Intel ones did. That may get some "bonus" time for Intel systems because the user base will still be full of them for a longer period of time. They probably won't sell poorly, but remains to be seen that they are going to radically change the consumer dynamics here. The price points probably will be about the same ( if apple does one for one replacements a few at a time. ). The folks who couldn't afford one before probably aren't going to buy one in the future if price levels stay the same.
These are going to be better Macs for good (to very good) upgrades for most folks but probably aren't going to cause the utter collapse of Windows market. There is probably some user base percentage cutoff where Apple would become more itchy trigger finger on getting out earlier. My guess is that is in the 10%-20% range left being on Intel for updates end and an expectation that will shrink to smaller single digits by the time the high impact security patches stop rolling out. That is probably pretty close to Apple's average 6 year to drop cycle. ( being real, systems were going to get dropped anyway at some point around 6-7 years with no ARM transition if Apple did an across the board rash of x86_64 updates in a single year. In part it is going to be the across the board update that going to get all the current systems dropped later in this decade. )
there are 100+ M x86_64 Macs out there. It is going to be kind of hard to make 80-90M of them disappear from active deployed use in a short number of years. Apple just doesn't sell that large amount in a single year to do that extremely fast and fair amount of these are going to escape out into the 2nd (or 3rd) owner and still actively used market.
Last edited:
Yup, my 2013 15” Haswell quad core i7 MBP with 16GB RAM and 1TB SSD is still running like a beast, probably still outperforming the 13” MBP of today, and definitely running circles around the MBA of today.
My 2017 iPad Pro is still humming along beautifully today and could last at least another 1-2 years.
It took 7 months in 2006 (from the MacBook Pro release on January 10 to the release of the Mac Pro on August 7).
PCIe already goes directly to the CPU. New Intel chips will have the Thunderbolt logic built directly into the CPU.
I'm aware AMD isn't doing it for their APU's but that doesn't mean Apple won't, they aren't AMD. Also in conversations like this I wouldn't really say things like "Err no" unless you have facts. It's all just opinion until they ship product, definitive statements can't be made.
I hope they’ll just release an upgrade board/option for 7,1 users. I don’t think they designed they case just for one revision.
Like the 6,1? Considering the transition time frame - this may be the case. Literally.
Looking at the development requirements for Big Sur on Apple Silicon, my Enterprise static content rendering app should compile without the need for any changes. Getting the hardware for $500 is nice.. But where is my macOS Simulator running in xcode? if I can get a simulator for an iPad Pro running ios. Getting the same for MacOS shouldn;t be far away.
FWIW.... The last macOS library I added to xcode was Google Firebase, an iOS cocoapod library that's cross compatible with macOS.
Should I be concerned with Big Sur compatibility? Not a chance. Going to 64-bit was much more work...
Last edited:
Power Mac G5 -> Mac Pro 1,1 didn't get a board upgrade option, even though they reused the same case. Apple can change a lot of the internals while the outside still stays the same.
No chance of that. The architecture will be too fundamentally different. There were 68K to PPC upgrade kits (from the last of the latter to the first of the former) because the PowerPC 601 processor Macs, in both hardware and software, were designed with backwards compatibility in mind- they had to be otherwise the OS wouldn't have worked. That, and Apple's dire financial straits at the time. You could turn your Centris 610 into a PM 6100/66 (for example) for this reason, though at great cost. Given that PowerPC and x86 were so different, it wasn't possible to do the same when the Intel switch came. Turning any Power Mac G5 into a Mac Pro would require gutting the case completely. It won't be any different this time. The 8,1 will almost certainly look the same as a 7,1, I doubt the internal layout will change much (unlike the G5 to 1,1 change), but that'll be it.
Anyhow, this is pretty interesting time. Unlike PPC to Intel transition, I’m not bound by OS anymore. For me, Windows 10 is plenty enough. Even better, the true work is done mostly on linux nowadays.
I’m really eager to see what Apple’s planning to do for Mac Pro sector. Until now, there’s no such indication that Apple’s dedicated GPU is in pipeline for example. What about Apple Silicone counterpart to Intel Xeon? That’s huge gap going from ipad pro chips to a truly behemoth scale chip with so many PCIe lanes, memory bandwidth and various functions.
As a geek, this whole adventure is really fun to watch.
What’s the worst thing that can happen?
- Apple suddenly offering gimped down, customer friendly Mac with Axx chips with integrated graphic only option, always sharing ram with GPU, more closed system, ios like mac that forces individuals to ios stuff. No more Mac Pro level professional Macs offering by Apple.
On the other hand, this may well be another stepping up by Apple with evetual release of true professional level Mac Pro like Arm workstation that will make all of us happy.
Either way, I will always have a back-up to fall back to. Can’t wait Ryzen4 with 16/32 3950x like CPU coming out in later 2020. Before Arm announcement, my upgrade plan for the desktop was 4950x with nVidia 3090 or ti. Well by end of 2020, the roadmap by Apple will be more clearer and, hopefully, comparable to what the other sides will offering.
I’m really eager to see what Apple’s planning to do for Mac Pro sector. Until now, there’s no such indication that Apple’s dedicated GPU is in pipeline for example. What about Apple Silicone counterpart to Intel Xeon? That’s huge gap going from ipad pro chips to a truly behemoth scale chip with so many PCIe lanes, memory bandwidth and various functions.
As a geek, this whole adventure is really fun to watch.
What’s the worst thing that can happen?
- Apple suddenly offering gimped down, customer friendly Mac with Axx chips with integrated graphic only option, always sharing ram with GPU, more closed system, ios like mac that forces individuals to ios stuff. No more Mac Pro level professional Macs offering by Apple.
On the other hand, this may well be another stepping up by Apple with evetual release of true professional level Mac Pro like Arm workstation that will make all of us happy.
Either way, I will always have a back-up to fall back to. Can’t wait Ryzen4 with 16/32 3950x like CPU coming out in later 2020. Before Arm announcement, my upgrade plan for the desktop was 4950x with nVidia 3090 or ti. Well by end of 2020, the roadmap by Apple will be more clearer and, hopefully, comparable to what the other sides will offering.