If they cut support with macOS Ventura, people will sue them. I would join that law suit. That's ridiculous.
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 Waiting for Mac Pro 8,1
- Thread starter Slash-2CPU
- Start date
- Sort by reaction score
There is no urgency for Apple to ditch the support of Intel-based Macs, unlike the period of PowerPC to Intel transition when G5 brought fatal troubles to Apple Computer. I still expect premium Intel-based Macs to get at least macOS 15 following the current pace.
Apple is probably not going to put this into the abnormally large category. The last Intel 27" iMac maxed out at 128GB
https://support.apple.com/kb/SP821?viewlocale=en_US&locale=en_US
1 TB of memory is 10x bigger than that. That would be a clear "abnormally" large winner.
The new M2 SoC leverages 12GB stacks of RAM dies. The Pro , Max , Ultra put two of those stacks into a single "container" package ( partially to cut down space). So an Ultra using same memory controller tech would have 16 stacks of 12GB . So that could easily scale up to 192GB for a "Ultra" sized SoC for a "lower half range" Mac Pro. So Apple would have 192GB covered easily just with the same tech using for M2 generation.
If there was a four die option that had "two Ultra" worth of memory controllers , then the max threshold moves to 384GB . Folks in the 128GB range could go to double working set size (256) and still have headroom left. The "upfront" costs are substantially higher , but they wouldn't be blocking folks from work who plan ahead.
That dedicated RAM SSD for compressed swap would work fine for this. Abandoned for 2-3 days apps. Those could be shoveled to secondary memory and the restart time would be relatively very short. Even if not there having more E cores around to do the compressing of long term, unused open apps would work also. Decent change if there were 6-8 E cores around that could get a smaller footprint due to macOS compressing those apps down more often.
With Apple moving to doing slightly bigger RAM packages they could do ECC with a modified memory controller. ( similar how GPUs tend to do it. )
If have set of 4 dies with 3GB each ( so 12GB ) , but use just one of those four as checksum extra then "usable" capacity drops to 9GB (and 3GB of checksum space).
So 16 die stacks of 9GB would be 144GB and 32 die stacks would be 288GB .
It isn't better than the current MP 2019 tops out at , but it is better than the final max RAM capacity for the iMac Pro ( 256GB ). Both cover the presented 128GB use case ( just not as much headroom). And the max end user available bandwidth would drop. [ Probably could see Apple doing some tap dancing where GPU RAM is segregated into memory controllers that don't take the bandwidth hit. But a slippery slope if doing enough tap dancing could just have two different RAM types with different latencies. A slippery slope not sure Apple is willing to go down. ]
Or Apple squeezes in another memory controller for each group of 4 (and an extra checksum stack) and no bandwidth hit. ( and perhaps goes down path Amazon did for Graviton 3 and push the memory controllers to another chiplet so can tolerate more fan-out for "extra" lanes. )
The root cause problem issue here is that iff look at the die of a Max or Ultra package the vast majority of that space and memory controller workload is being thrown at GPU cores. Apples make driver is mostly on the GPU part of the subsystem. And generally GPU workloads don't give a crap about memory accuracy. The hype train is all about frame rates per second. The frame with the 1bit error is thrown away faster than you can blink. So basically nobody cares.
Nvidia is doing it on their Grace SoC using LPDDR5. But at the price points Apple wants to deliver at? Unclear.
Yea cost point is a really good question. These chips end up being ONLY for low volume Mac pro's with special controllers? Maybe. It will be interesting to see what they do.
Six months ago I was optimistic about x86-64 macOS support, as optimistic as towards the end of this decade (counting in extended years of security updates). That was based on assumption of Apple being cautious about Mac Pro transition. Includes introducing AS Mac Pro based on M1 SoCs, and a refresh of MacPro7,1 with e.g. processor update (to co-exist M1 MP for a period of time). Both didn't happen.
Only remaining Intel Mac's are Mac Mini 2018 and Mac Pro 2019. Seems more and more likely Apple will be able to replace and discontinue both within 12 months.
M1 Mac Mini lacks memory coverage. I think that's the main reason Intel Mac Mini (max 64GB RAM) not killed yet. With M2 SoCs and high density LPDDR5 chips, Mac Mini with M2 can be made to support max 32GB memory. If Apple further introduces a Mac Mini model with M2 Pro, then max memory will reach 64GB. Intel Mac Mini... finally done.
Apple discrete GPUs are more likely than people (at least myself lol) initially thought. This may be the biggest change of thought in the last six months. I also thought DIMMs a possibility for a long while. Seems more people tend to agree now. With both plus a few PCIe slots, when the new AS Mac Pro launches, I think Mac Pro 2019 will be discontinued.
I feel a very confident Apple... pretty confident about its new Mac Pro. Seeing how M2 MacBook Air looks, people can be assured the new AS Mac Pro will look more like Cheese Grater than Trashcan.
When people at Apple Spaceship think no urge to stop x86-64 support, they may include x86-64 in macOS 14. Otherwise, possibly macOS 13 is the last. But as I said in my last post, support of MacOS 13 i.e. security updates & small bug fixes for x86-64 could be extended from two to four years. That gives about 2017. So probably not that hard for users to swallow Apple would think.
The choice in withdrawing support this year is a bit odd versus years where it was pretty consistent (requiring Metal, etc.), and I'd be curious what the reasoning is for some of them (at least with the MBPs, maybe something to do with the keyboard warranty stuff, who knows?) At least for the 6,1 Mac Pro, it is a 2013 product; anyone who bought one because they needed one years after the fact knew what they were getting into in that respect.
The last PowerMac G5 got three full years of support (and then the security updates after that) after it was discontinued. I'm pretty sure the 7,1 will get about that as well. Beyond that? I'm sure the enterprising people who get the newest OS working on the 5,1s will be able to work their magic as long as there's an Intel version of macOS made (hell, and possibly even after.) If you don't need official support there's plenty of avenues available to you. Not like computer makers bust into your house and hijack your stuff when it's officially obsoleted.
From a business standpoint, if you can't make a professional product worth the money after only a few years, I don't think you're investing wisely in equipment. The calculus is a lot different for a regular consumer, but with the high starting price I feel like people should read the writing on the wall and remember that it's a high-end product, and given the pricing of the Mac Studio the 8,1 will likely maintain that. (I could see an entry-level model at $4999, instead of $5999, but it's still going to be super-pricey.)
LPDDR5 (and DDR5) already have built-in ECC inside the RAM chips. Any single bit error is automatically corrected. Apple silicon SoCs (and its built-in LPDDR5 memory controller) are in very close proximity to LPDDR5 chips. The traditional sense of ECCs in my opinion won't be required as the signals travelling in-between are v. v. v. close. I think that's the main reason e.g. M1 Ultra with 128GB RAM the genius at Apple Spaceship doesn't lose any sleep at night.
If AS Mac Pro comes with DDR5 DIMMs (in hundred GBs up to a few TBs), of course, the traditional sense of ECC memory will be mandatory. So perhaps only the SoC for AS Mac Pro will have its memory controller with ECC built-in. These will be additional memory controllers anyway to handle DDR5 DIMMs because the other controllers will be still required to handle LPDDR5 memory for GPU (and CPU). BTW, I believe we're going to see Apple's simple yet innovative use of memories here.
If AS Mac Pro comes with DDR5 DIMMs (in hundred GBs up to a few TBs), of course, the traditional sense of ECC memory will be mandatory. So perhaps only the SoC for AS Mac Pro will have its memory controller with ECC built-in. These will be additional memory controllers anyway to handle DDR5 DIMMs because the other controllers will be still required to handle LPDDR5 memory for GPU (and CPU). BTW, I believe we're going to see Apple's simple yet innovative use of memories here.
I dont know where you get on with this macOS 13 is the last intel OS. First, it would be a COMPLETE departure from their past. They usually provide support for at least 3 versions of the OS after dumping. They are still selling intel machines. So we're talking about macOS16 if you go with the idea that they kill off the 7,1 around December of this year. Second, there would be a howls and lawsuits. And no, simple security and bug fixes dont count. And third, the scale of intel machines is orders of magnitude greater installed base than powerPC machines had; a LOT more people would be pissed. It's like youre trying to start a rumor that is counter to all apple precedent on support.
Apple introduced intel Macs in 2006 and the last powerpc macOS was lion, which was the OS from 2011 and not replaced until July of 2012. That's 5-6 years depending on how you count.
The M1 Mac was first released in November of 2020. This transition is taking apple longer than the 2 years, but by even the most aggressive count, you'd expect support at least that same 5 years, so through November of 2025. So Ventura macOS 13 will be out in 2022, that leaves macOS14 for 2023, macOS 15 for 2024, and macOS 16 for 2025.
Last edited:
My position hasn't changed in the past few posts i.e. MacOS 13 is possibly the last or second last x86_64 MacOS release. If it's second last, then MacOS 14 may be the last one. I don't know about others saying MacOS 13 will be the last. But my reasoning is based on some aspect of PowerPC to Intel transition.
This doesn't sound correct to me.
PowerPC to Intel transition finished very rapidly:
- migration from PowerPC to Intel announced in 2005, and intended to finish by 2007
- product release of Intel Mac started in early 2006, and finished by the end of 2006
- last PowerPC MacOS (Leopard) released in Fall 2007*; last update to Leopard Fall 2009. No more OS update for PowerPC after that.
Fast forward to now.
If last member of Intel Mac's will be migrated by end of 2022, MacOS 13 (to be released in Fall 2022) is likely the last x86_64 MacOS. If Apple thinks ..let's be nice.. then MacOS 14 may the last x86_64 release.
There are way more Intel Mac's and still capable and useful. To sweeten the bitterness, I was suggesting Apple may extend update of last x86_64 MacOS from the usual two years to four. Let me spell out the letters for you by an example, say, MacOS 13 the last x86_64 release:
- MacOS 13 to be released in Fall 2022
- one year of active update and bug fix to end in Fall 2023
- usual two years of security updates to end in Fall 2025
- two extra years of security updates to end in Fall 2027
and if there is an 2022 mac pro X86_64 cpu / video card bump?
Will mac os end rosetta in MacOS 14? if they also end x86_64?
Will they keep rosetta and x86_64 around till they end rosetta
Wrong.
Intel Macs released 2006. Who cares when they were announced.
"The Intel-based iMac is a discontinued series of Macintosh desktop computers designed, manufactured and sold by Apple Inc. from 2006 to 2022."
iMac (Intel-based) - Wikipedia
Last powerPC macOS was lion and released in 2011 (and not replaced till 2012).
"Lion was released to manufacturing on July 1, 2011"
OS X Lion - Wikipedia
Do the math. It's 5 years from release of the first intel Mac while previous architecture was still maintained (dropped in the 6th year by mountain lion 10.8).
So intel machines might still be sold in 2023 (because they are running late on the 2 year conversion), but the last supporting OS is macOS 13, and 'youre doing us a favor' by saying macOS 14. So don't do us any favors. Have the guts to say what you think, which is, macOS 13 will be the last intel OS released in 2022, and intel machines might be sold, straight into 2023. Sure sure. 🙄
On what basis would sue them? I am not aware of any contract binding them to support existing system with new operating system versions.
IMO any professional who didn't learn from the 6,1 Mac Pro fiasco shouldn't be critical of Apple if they drop support for their existing systems.
Many bases. With all their domestic and foreign antitrust woes, doing this would be a horrible look for them, and people would use antitrust as well. Ooh, and false representations (they said when the transition was starting that intel machines would be supported for a long time). Apple just paid a settlement to Chile for planned obsolesce. Apple settled for 500m for the apple slowdown/battery misrepresentation debacle. That would look like sunshine compared to dropping intel hardware with macOS 14 by way of misrepresentations. Oh yea, and the GLORIOUS discovery on the planned obsolesce treasure trove PR hit would be way in the billions for them wrt share price hit and reputation hit. It would be a bloodbath of damages to apple.
Are you kidding. The 6,1 is a 2013 machine and it was supported with new operating systems through macOS12. Basically 9 years of releases. It was well supported by way of operating system releases.
Rosetta was created for PowerPC to Intel transition, enabling execution of PowerPC binaries on Intel Mac.
Rosetta 2 is for Intel to ARM transition, for executing x86_64 binaries on ARM Mac.
N.B. Don't confuse Rosetta as PowerPC MacOS release or Rosetta2 as x86_64 MacOS release. Apparently some people in this thread did and still do. LOL
Both software can theoretically live on in subsequent MacOS releases forever. There was little incentive to keep Rosetta because PowerPC had been lame ducks even before being replaced by Intel processors.
x86_64 is still the performant and dominant ISA. Apple has little reason to drop Rosetta 2 soon, and perhaps they can let it live on as long as x86_64 is dominating the world.
***
Intel refresh for MacPro7,1 still possible? Now is mid 2022. I think it's less and less likely.
GPU refresh possible? Perhaps. We can look for hints in MacOS 13 kexts in the coming 12 months. If MacPro7,1 is discontinued by Fall 2022, and there is no Intel refresh to MacPro7,1, it also seems not logical to expect standalone release of new MPX GPUs.
OK, what are they?
Which they only stopped selling a little of two years ago.
You are mistaken.Last powerPC macOS was lion and released in 2011 (and not replaced till 2012).
"Lion was released to manufacturing on July 1, 2011"
OS X Lion - Wikipedia
en.wikipedia.org
The last version of Mac OS X to support PowerPC was Leopard (10.5), released in 2007. The last security update for Leopard was released in May 2012, I believe.
Snow Leopard (10.6, 2009) was Intel only.
Last edited:
I agree, if Apple wants to make the transition from x86 > ARM as smooth as possible, and entice as many unwilling customers (such as myself), then they will absolutely need to keep Rosetta 2 going for at least 2 more years, if not longer, and like you said -- it wouldn't make sense to completely abandon it, at least not while software on ARM is limited (and I imagine at the enterprise level even more so).
You'll get no arguments from me, but the other day, I found THIS very interesting site, some food for thought...
Carry on!
Last edited:
What are you not understanding. False representations are actionable. Antitrust is actionable. They have been sued and lost for misrepresentation as I noted. They are currently being sued for antitrust, so things like planned obselecnse are actionable. Just actually read.
I accept your apology.
You are correct and I stand corrected. 10.5 was the the last version. There is a good chart here:
macOS version history - Wikipedia
And I stand corrected on my 5 year assertion then as well! Intel machines released on 2006, and the east OS to support intel machines was released October of 2007. So 1 year after.
As such, @kvic was right, and I was wrong, and macOS14 could/would be the last Intel update we can expect if the past is any guide to the future (and it often is).
Why you refuse to provide the basis upon which you would bring a lawsuit. Saying "false representation is actionable" sounds good but it needs to be backed with an actual false representation. I am not saying that one doesn't exist but rather I am unaware of it so I have asked you to provide it.
IMO antitrust is irrelevant to this discussion.
I offered no apology so there was none for you to accept.
Last edited:
Your ignorance of the law is astounding and limitless. I just pointed out apple was SUCESSFULLY sued for product misrepresentation with regard to battery features and planned obsolesce. They represented during the initial keynote they would keep intel machines supported for a long time. If they do not, that's a misrepresentation for which they can be sued. Also, that it's planned obsecelence, meaning they obsoleted something purposefully that needn't be obsolete.
Those are things apple has LOST on.
That you say antitrust is irrelevant is your misguided opinion that you are fully entitled to.
What Apple was successfully sued for is irrelevant. I want to know what misrepresentation you allege would be the basis of the lawsuit you referenced.
Instead of providing this basic information all you can do is hurl ad homs. Instead of continuing the insults how about just answering the question? There's no need for hostilities.
Um you CLEARLY do not understand how law works. Successful suit sets precedents. So when you lose for planned obsolesces once, you likely will lose when youve done it again. Also, the misrepresentation is that they promised, when they announced the M1, that they would keep intel around for many years to come. If they kill it 1 year after the last intel machine is sold, that will be a material misrepresentation.
Last edited:
LPDDR5 and DDR5 do not approach ECC the same way. Nor are their standard bus widths the same (which partially drives the different choices ).
"... As DDR5 and LPDDR5 support much higher data-rates than their predecessors, they support additional ECC features for enhancing the robustness of the memory subsystem. On-die ECC in DDR5 and Link-ECC in LPDDR5 are two such RAS schemes to further bolster the memory subsystem RAS capabilities. .."
https://www.synopsys.com/designware-ip/technical-bulletin/error-correction-code-ddr.html
There is no default "built in" ECC for LPDDR5. The standard LPDDR5 is aimed protecting the amphoral transfers of data on the bus/channel from the Memory package to the SoC package. When the data is at rest it doesn't protect anything. ( that saves a substantial amount of capacity as a trade off. )
DDR5 is the one that has "at rest" requirement turned on all the time. The link protection is optional and delivered via a wider side channel.
LPDDR5 can do it without a side channel, but that will take a bandwidth hit as the ECC data has to trail behind the data all the time on a relatively narrow channel ( not travel in parallel. ) . Seems unlikely Apple is going to toss data bandwidth out the window.
Apple is pragmatically running a "poor man's HBM" memory solution using LPDDR building blocks. So they cluster 4 LPDDR buses together and aim them at the same RAM die stack . They could 'weave' the parallelism in that subsystem to avoid the inline bandwidth slow down depending upon how they are clustering together the multiple die read/writes ( or not ) . Apple is already running a high rate of die access concurrency into a single stack so won't raise power dissipation issues of the stack much with ECC date traffic weaved in.
Apple isn't using standard , off-the-shelf LPDDR. The basic materials are the same (so don't need different RAM dies) , but how they are stacked and connected isn't the "norm". Especially, those chasing higher capacity at lower costs (via more banks and a fair amount of non-concurrent die access. ) It is different than the norm but Apple asks just one or two makers to make very high volume of it ( basically the whole line up using the same basic building blocks with some minor differences on packaging for the Pro/Max to save more space. Two side-by-side die stacks in those packages so easier to pack them in around the sides than one-stack-per-package plain M. ) .
Actually not traditional ECC. DDR5 still can be run without the link protection to/from the package. There is explicitly optional ECC DIMMs to do both. Still such a think as "DDR5 " and "DDR5 ECC" distinction.
As for AS Mac Pro ... not really likely. DDR5 isn't oriented toward doing the "poor man's HBM" solution that Apple is hooked into. It is a wider bus so not going to as compactly fit a relatively high number of DDR5 channels into a small space. Not impossible to do, but harder to do than with the narrower LPDDR5.
There are really good reasons why don't see high end GPUs using DDR5. The number of cores making the number of pragmatically more random requests means more reasonable size paths/channels is better than fewer wider paths.
So perhaps only the SoC for AS Mac Pro will have its memory controller with ECC built-in. These will be additional memory controllers anyway to handle DDR5 DIMMs because the other controllers will be still required to handle LPDDR5 memory for GPU (and CPU). BTW, I believe we're going to see Apple's simple yet innovative use of memories here.
Seriously not the practice that Apple has engaged in so far. They use the same blocks of 4 channel LPPDR memory controller and just replicated it into bigger aggregations up the whole food chain. From A-seires up to Max.
The overhead of doing it is spread over all the implementations.
It would be surprising if it went onto the primary die though. Intel has done lots of work to weave Optane DIMMS into the same space are DDR DIMMs. Mismtached memory types would leak into the operating system kernel infrastructure also.
Apple doing something "Simple" and "cost effective bill of materials" would be doing homogenous memory.