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
I bought one of the last generation PowerPCs and it sucked and I abandoned Apple for a while over it. I spent a lot of money just out of college and it was basically useless for the video editing I bought it for within a couple years.
Avid stopped supporting it quickly was the big issue for me. And two years after I purchased it they stopped offering OS upgrades.
I am a hobby user now with a fair amount of expendable cash and I am glad I didn't invest in one of the Mac Pros. It will be a 500 lb paperweight in a couple years.
And the idea that these computers should just be useless in 5 years is laughable. My 2009 Mac Pro that I bought used in 2014 lasted me years and now my father in law is using it. Even the 2013 models are still going strong. The people who bought the new Mac Pros are not going to have any type of lifespan like that.
Avid stopped supporting it quickly was the big issue for me. And two years after I purchased it they stopped offering OS upgrades.
I am a hobby user now with a fair amount of expendable cash and I am glad I didn't invest in one of the Mac Pros. It will be a 500 lb paperweight in a couple years.
And the idea that these computers should just be useless in 5 years is laughable. My 2009 Mac Pro that I bought used in 2014 lasted me years and now my father in law is using it. Even the 2013 models are still going strong. The people who bought the new Mac Pros are not going to have any type of lifespan like that.
the sake of backward compatibility just holds back innovation, it's about time to ditch x86, Windows, & macOS.
we milked them long enough, small improvements (not always, but in general) but still all the old load on board.
imo its time to leapfrog forward, ARM is the enabler, Apple has a chance to revolutionize the entire industry with this.
and to be honest, that is what Apple wants more than anything else, they will try for sure.
Microsoft with surface X aint changing the industry, but it tells where we are headed to, and how not to do it.
Amazon, Google, Apple, Microsoft, they all want ARM so badly, the switch is already ongoing behind the scenes.
I don't see a path for Intel or AMD really. after failing 10 years in a row, AMD has catched up to Intel, better in some areas but generally still inferior, but now the Price/Performance is a killer argument for Customers.
this let Intel "sleep" and deliver only minor improvements over the Years while charging Big numbers, simply because they could.
meanwhile the smartphone took over the World,
Usercount is through the roof, a highly competitive an incredibly rewarding Sector
that led to massive Innovation.
and its all based on ARM.
Amazon, Google, Apple, Microsoft, they all want ARM so badly, it's inevitable at this point.
what this means for the Mac pro ?
depends how far in the future we look but in a Decade
it could have 1000 cores without active cooling and fit in a Mac mini enclosure or even your Pocket.
Today the Ampere Altra Server CPU has 80 Cores, is dual socket capable (160 cores), 4TB RAM per Socket = 8 TB RAM 3Ghz per Core, 210 Watt TDP
and honestly Apple is way better at this than Ampere, Apple has proven over and over again to make the best Chip every Year over a decade now.
the first ARM based Mac pro could have 200-400 cores
we milked them long enough, small improvements (not always, but in general) but still all the old load on board.
imo its time to leapfrog forward, ARM is the enabler, Apple has a chance to revolutionize the entire industry with this.
and to be honest, that is what Apple wants more than anything else, they will try for sure.
Microsoft with surface X aint changing the industry, but it tells where we are headed to, and how not to do it.
Amazon, Google, Apple, Microsoft, they all want ARM so badly, the switch is already ongoing behind the scenes.
I don't see a path for Intel or AMD really. after failing 10 years in a row, AMD has catched up to Intel, better in some areas but generally still inferior, but now the Price/Performance is a killer argument for Customers.
this let Intel "sleep" and deliver only minor improvements over the Years while charging Big numbers, simply because they could.
meanwhile the smartphone took over the World,
Usercount is through the roof, a highly competitive an incredibly rewarding Sector
that led to massive Innovation.
and its all based on ARM.
Amazon, Google, Apple, Microsoft, they all want ARM so badly, it's inevitable at this point.
what this means for the Mac pro ?
depends how far in the future we look but in a Decade
it could have 1000 cores without active cooling and fit in a Mac mini enclosure or even your Pocket.
Today the Ampere Altra Server CPU has 80 Cores, is dual socket capable (160 cores), 4TB RAM per Socket = 8 TB RAM 3Ghz per Core, 210 Watt TDP
and honestly Apple is way better at this than Ampere, Apple has proven over and over again to make the best Chip every Year over a decade now.
the first ARM based Mac pro could have 200-400 cores
I wouldn’t be surprised if Apple puts up a slide at WWDC saying that the year that Intel Macs will be supported with macOS updates is...
2030.
So go ahead and buy that new Intel iMac.
Although I wouldn’t be surprised if after 6-8 years there’s a feature freezed version akin to Snow Leopard for Intel Macs that only receives security, Cloud compatibility and any hardware (glasses, watch etc) updates.
2030.
So go ahead and buy that new Intel iMac.
Although I wouldn’t be surprised if after 6-8 years there’s a feature freezed version akin to Snow Leopard for Intel Macs that only receives security, Cloud compatibility and any hardware (glasses, watch etc) updates.
ARM on a Mac Pro actually has a potential of being a beast. Professional software heavily relies on vector instructions and CPUs that can process multiple data points at once. For example, Intel's newest vector instruction set is AVX-512, and CPUs that support it can process 16 numbers in one go. To harness this power however, the software needs to support AVX-512 specifically. And of course, since earlier CPUs don't support it, you also probably want the software to support AVX2 (which is a very different thing). All of this is rather messy and so it takes years until the true potential of the new CPUs can be unlocked.
But ARM has this thing called SVE (Scalable Vector Extensions). With SVE, the code will automatically adapt to what the CPU can do. Say you want to add one thousand numbers together. When you are using SVE, the CPU will automatically process the data in appropriately sized chunks. An iPhone CPU for example can only do 4 numbers at a time, so it will take 250 steps to complete the task. But a more powerful ARM CPU could do 16 numbers at a time (just like what Intel does) — needing only 63 steps. And this all with the same identical code — no need to rewrite the program to take advantage of newer features, no need to wait and test the tooling. Say two years later a new CPU comes up with 1024 wide registers — you just start your software and your performance in these areas has doubled. This is currently totally impossible with Intel's approach that requires you to implement a whole new technology first.
Personally, I think this is extremely exiting and I believe it will unlock a completely new horizons for pro appliations (and not only). Say, you can write some ML procedure once and have it run on all classes of hardware — mobile, desktop, pro desktop, supercomputer. No need to tweak, optimize and test it on all the architectures, and all that developer time and money can be used to implement more features or make sure that the code is higher quality of what would have been possible otherwise (of course, some tweaking is inevitable but the amount of work is significantly reduced).
But ARM has this thing called SVE (Scalable Vector Extensions). With SVE, the code will automatically adapt to what the CPU can do. Say you want to add one thousand numbers together. When you are using SVE, the CPU will automatically process the data in appropriately sized chunks. An iPhone CPU for example can only do 4 numbers at a time, so it will take 250 steps to complete the task. But a more powerful ARM CPU could do 16 numbers at a time (just like what Intel does) — needing only 63 steps. And this all with the same identical code — no need to rewrite the program to take advantage of newer features, no need to wait and test the tooling. Say two years later a new CPU comes up with 1024 wide registers — you just start your software and your performance in these areas has doubled. This is currently totally impossible with Intel's approach that requires you to implement a whole new technology first.
Personally, I think this is extremely exiting and I believe it will unlock a completely new horizons for pro appliations (and not only). Say, you can write some ML procedure once and have it run on all classes of hardware — mobile, desktop, pro desktop, supercomputer. No need to tweak, optimize and test it on all the architectures, and all that developer time and money can be used to implement more features or make sure that the code is higher quality of what would have been possible otherwise (of course, some tweaking is inevitable but the amount of work is significantly reduced).
Surely, this is not something Apple would taking lightly. I wouldn’t be surprised if they’re not already working with some large developer. They won’t flip the switch until they’re sure that it’ll work technically.
My concern would be over confusion it may create with customers during the transition. But again, Apple would be wise to make the switch over gradual starting with a MBA and/or MacMini.
And sure, they’re a much bigger company now when they were back then, but this is their Macintosh user base, not their iOS. It’s much much smaller.
I wouldn’t be too concerned about the support life of the current MacPro or any other intel based Mac either. We live in a very different world to the one at the time of the intel transition where security is much more important and devices last for a lot longer in service. I’d also not be surprised if we get another Intel refresh later this year with the first arm devices shopping next year in limited quantities only. I think we will see OS X versions for at least 3 more years on intel and security patches for at least 3 post that.
As for the is it a good thing I have no strong opinion, certainly in DCs and on workstations we don’t really see any major breakthrough yet although companies like Amazon are throwing serious money behind it so it might not be that far away. It’s a big gamble, I wouldn’t be surprised if Intel pull something out the bag that keeps them competitive, unlike previous chip suppliers they have a huge market and masses of cash to throw at the problem
As for the is it a good thing I have no strong opinion, certainly in DCs and on workstations we don’t really see any major breakthrough yet although companies like Amazon are throwing serious money behind it so it might not be that far away. It’s a big gamble, I wouldn’t be surprised if Intel pull something out the bag that keeps them competitive, unlike previous chip suppliers they have a huge market and masses of cash to throw at the problem
Licensing rights aren't really the big blocker. A lot of folks just didn't want to pay Intel anything more than had to pay in the x86 space anyway. ( including AMD which has chunks of x86-64 that they own but have a cross licensing with Intel on. ), Obviously Apple wasn't one of those folks.
It is more so doing the work. At one point Apple trademarked "Thunderbolt" name. but they handed that plus the work they did on the copper (as opposed to optical ) Thunderbolt over to Intel to build out as a product standard. Apple may have been a small piece of the action ( and/or a larger discount on TB controllers) that flowed through the royalties, but it is owned by Intel. ( Apple largely being the biggest buyer and throwing around meant they had influence though. )
Intel has made the basic protocol tech paid royalty free and had gotten USB to adopt large parts of it for USB 4 and the DisplayPort folks to adopt a variation of the implementation for DP v2.0. Apple has been involved along the way ( e.g., worked with Intel to put substantial resources into getting USB Type-C put together. Intel and Apple were the biggest contributors to that project. )
Calling something "Thunderbolt" still has to go through an Intel litmus and compliance test (the brand Thunderbolt they didn't have over). Technically what most folks would commonly call "Thunderbolt 3" isn't required by USB 4. It is optional in many significant places in the standard ( e.g., on host systems. one port peripherals Both can claim to be USB 4 and offer no TBv3 compatibility but still be using some basic TB protocols to transfer data in a somewhat different implementation. ) . That is a better adoption than nothing but there are still folks in USB-IF that probably would like to kill off Thunderbolt. Little good sense in handing over the name too. USB really doesn't want the 'name' itself.
there are still no shipping USB 4 discrete controller vendors. Supposedly Asmedia may have something by end of this year. ( as story from last year).
USB-IF: USB4 Products Coming in Late 2020
No Computek show at the end of May this year so not much publicly visible indications they are still on track. Intel should have a discrete controller too out by end of the year but that doesn't have much public presence. ( but Tiger Lake appears to have USB4 with TBv3 present so it would need to hook to something to be a value add.)
It appears that "Thunderbolt 4" is pragmatically more an indication that the "optional" TBv3 parts of USB4 are actually present and implemented well. ( passes test and validated. ). That would still be controlled by Intel
Your metrics are off quite a bit here. These are old numbers. Before Moore's Law became irrelevant. Before The Plataeu.
in 2020, I'd say it's responsible to consider assessing a new system at 5 years, but Macs are usable for a considerably longer period of time. Anything sooner and you're probably just flushing money down a toilet. I work professionally in print on a 2013 running Mojave. I run a 2015 at home. Performance is the same as it ever was on both. It's pretty safe to assume you can get 7-10 years out of any recent Mac. No question.
Apple probably wouldn't even be supporting a 2021 ARM Macbook in 2030. If got superceded in two years then 2023 would be end date. 6 years after that is 2029. Most likely it would end there. Only get 5-7 years after last sale for support.
Obtaining service for your Apple product after an expired warranty - Apple Support
Learn about your options for getting service and parts for Apple devices that are past their warranty period.
Most likely something in middle to bottom of the range gets about 6 year ( middle of their 5-7 range) and done for official OS upgrades after hardware window dries up.
If Apple hits the snooze button on the 2019 Mac pro for 4 years then could end up in 2023 also with either a discontinue or replacement.
Your scenario seems probable. The neglect of Macs and macOS in particular have pointed to an impending shift for a long time. I think they got the timing wrong and had to turn out some belated updates as place holders but their heart hasn't been with the Mac for a long time.
A super-duper iPad can replace 90% of what people use general purpose machines for today. It will start "good enough" and work it's way up. There are a few years left for those who bought Intel machines recently. I would not make any long term plans with the platform.
Just as another data point: the MacOS kernel source code has included varying (increasing) degrees of ARM support since at least 2016, so they've had their eye on this for some time. Producing an ARM laptop/desktop that runs MacOS won't be a challenge for Apple - the challenge will be making the transition from legacy software palatable (because it certainly won't be painless). I won't hazard a guess about what tack they'll take, but there are some intriguing bread crumbs in the kernel source that suggest perhaps an emulator isn't out of the question...
You sound awfully certain the demise of the 7,1 will be a gruesome and painfully expensive death for its owners. What makes you so confident?
See, this is where I disagree. If there is to be a switch, they have to communicate clearly that the switch will happen across the board and provide a timeline of no more than a year for it to happen.
Switching gradually and with only certain products is problematic for two reasons: (1) it signals to the developers that Apple is not confident in the switch; (2) it gives permission for developers to take a wait-and-see approach because Intel will continue to be supported on new Pro hardware and thus even Apple must continue to support the Intel-based APIs.
Rather, I think the switch has to be total over a reasonable timeline to signal to the developers that they're either with Apple or they're out of the macOS ecosystem. Decide now, no opportunity to wait-and-see, because Apple is definitely going ahead with it.
This Twitter thread goes into a nicely detailed description of what such a switch entails, and why it must be absolute for it to work.
Thread by @stevesi: 1/ Newsflash: Apple going to bring ARM to chips to Mac. // What could this mean? Seems this must mean the best of all worlds. Is that how te…
Thread by @stevesi: 1/ Newsflash: Apple going to bring ARM to chips to Mac. // What could this mean? Seems this must mean the best of all wos that how tech transitions happen? Thoughts, scars… tl;dr not so simple or obvious. Reminder: Think Different. b…
History. Once they switch chips, all the software companies will start switching and will stop supporting Intel. They might have some emulator that will make software optimized for ARM run slow on Intel and call it "supported" and then they will just stop after a year.
Someone put the timeline up on this thread for what happened with PowerPC.
Within 3 years those top of the line professional computers you spent 20k on will be giant turds that won't run anything. They may "support" the OS for 5 years (but cut off upgrades after 2) but even still, all the software companies will move on quickly...so you can either run old versions or you will have 10k internet surfing machines.
Will come back here and laugh when the "1000 core Mac Pros" never materialize and it is shown that it was all a money grab and not some breathtaking innovation.
If this holds up and they exponentially surpass intel on raw performance in a Mac Pro (at least for a time), it could likely be powerful enough to emulate x86 (remember virtual PC?) and still perform at least equally to whatever current Xeons it will be competing against, even after the overhead.
Since the average age of hackintosh users is 72, that's just about right...
I am puzzled where such an idea originates from. I read all of the comments from the front page announcement and I was surprised with the number of comments to the effect ARM is the panacea for processor performance.
ARM is essentially a specification (similar to SPARC). How any given processor based on the specification performs boils down to implementation. I have no doubt Apple has some very talented chip designers and has done some excellent work implementing ARM based processors. However I doubt they're exponentially more capable than a company who's primary revenue is derived from manufacturing computer processors. This is not to say an Apple ARM implementation cannot outperform x64 but the performance claims, such as potentially being able to exponentially surpass Intel, overly optimistic.
Could these new chips be the stepping stone for Apple to make a serious jump into gaming? Create their own console and compete with PlayStation and Xbox!
Exciting times! I can dream!
Exciting times! I can dream!
I think he didn't mean to use exponential. I'm expecting Apple to deliver an IPC (Instructions Per Clock) increase over Intel by about 25%. But I'm also expecting Apple to deliver lower clock speeds than Intel which would allow Intel to claw back some ground at the expense of heat and power.
The thing is of course those are in short supply in the sorts of computers Apple wants to make which are thin and light portables. This is why we keep seeing Intel CPU's launch that have for example 5GHz turbo speeds but 3.2-3.6GHz all-core base clocks and when under load in Apples computers they can fall even below that due to thermal throttling.
I'm definitely not expecting Apple to deliver a core-to-core doubling of performance over Intel but I am expecting them to be able to deliver more cores than Intel can and with an IPC advantage so the cores Apple does deliver will do more.
In the near term Intel is going to launch a 38 core XEON processor appropriate for a Mac Pro refresh. I do believe Apple could deliver a 64 core chip where each individual core is 20% faster than those cores on the XEON. Intel is floundering with their 10nm fabrication process still unable to produce large dies and even small (4-8 Core) dies at high clock speeds.
AMD seemingly has no such problem and is already today as we converse here delivering 64 core processors with 128 PCIe 4.0 lanes and 8 channel memory interfaces. Intel is providing to the market right now today half the cores, 1/3rd the PCIe bandwidth and 85% the memory bandwidth. All while doing so at twice the cost that AMD is.
As great as Apples SoC's are currently they are up against a beaten opponent in Intel due to a fabrication roadmap that has gone completely off the rails. Essentially Intel has designs, great designs even. They just have no way to fabricate them.
buying computers that expensive as a personal computer doesn't seem logical anymore. I've stuck with iMacs the last few years which I replace every 4 years or so... though to be honest I'm not wowed by my 2017. I got the highest specced one and it doesn't feel any faster than my 2013 albeit with retina.... and at my office they got us top of the line 2019 ones and I'm still not thinking they're much faster than the 2017... I'm kinda hoping arm chips see greater speed increases between generations as time goes on
Perhaps not however I stand by my statement that many posters appear to hold the position that an Apple ARM implementation is the panacea for processor performance.
I consider this a temporary issue, especially given AMD is already fabricating on 10nm. Frankly I'm surprised Intel is having such difficulty. It's puzzling.
I think it's important not to lose sight that some of the issues Intel is experiencing aren't affecting AMD. While Apple only uses Intel processors we need to keep in mind they're moving from x64 to ARM, not Intel to ARM.
Microchips are very complicated and not many people will understand the nuances. I wasn't at any point disputing what you said about this. The ARM, SPARC, Power, RISCV, MIPS, x86 and others are only as good as their implementation which is a combination of design architecture and fabrication prowess.
Just merely changing from x86 to ARM based processors is inconsequential for me, it's all about the actual chips Apple will produce. That's what I care about and what I think all my comments in this thread have focused on, Apples prowess at making chips regardless of the architecture in use etc
Keep in mind Intel once had a 2 year lead on fabrication technology which has now fallen to a 2 year lag. Essentially 10nm is late by 4 years already, I wouldn't call that temporary and they've already pushed back XEON processors that were meant to launch this year on 10nm to next year.
TSMC is leading and it's who everyone uses. Qualcomm, NVIDIA, Apple, AMD etc. There's a lot of money going to TSMC and Intel is well just fabricating their own stuff, inherently risky.
Also I wanted to add Intels 14nm was late by two years also. Intel was meant to be on 5nm by now, currently their high performance stuff is on 14nm and they have a few niche low-power parts on 10nm (very small dies where high defects rates aren't so catastrophic due to how many dies they can produce per wafer even if 50% of them are duds).
Last edited: