Dead and buried?
I hope not, but it's quite possible.It's coming!
Apple said so. Guesstimates have been fall 2022 and spring 2023.
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.
Apple's Upcoming M2 Pro Chip for High-End MacBook Pro and Mac Mini Will Reportedly Be 3nm
- Thread starter MacRumors
- Start date
- Sort by reaction score
I hope not, but it's quite possible. Apple said so. Guesstimates have been fall 2022 and spring 2023....the M1 Pro starts at 16GB. As I posted in another thread, the difference between a M1 Pro MBP and M1 Max MBP is $600, so $1400 would be quite plausible for a $2000 Studio Max "downgraded" to M1 Pro.
The difference between Studio and Mini is then a pound or so of cooling hardware and a couple of front-mounted ports.
Seeing as has how Apple has dorked me over by effectively end of lifing devices like my 2018 iPad Pro, which can’t run Stage Manager, I’m in no rush to buy anything with an M2 chip — pro extreme or otherwise. I have no confidence that chips even a year or two old will be able (read allowed) to run newer system software processes or services. Can’t wait to see what my M1 Mac mini won’t be allowed to do.
Last edited:
What a mess, so much for simplicity. However, I hope that means a new iMac in late 2022 or early 2023 -- my late 2013 rMBP 13’ is on its last legs…
Last edited:
Quite frankly, I think Pro, Max, and Ultra are confusing on their own. Apple has an iPhone with the tag "Pro Max".
I'm not criticizing Apple for releasing a 3nm chip, I AM criticizing them for the very likely dumb slides which they'll show off at the next event comparing it to M2. They're basically two different chips with the same name. And how does that effect compatibility long term? 8 years down the road will they drop support for the M2 generation all at once or just cut off 5nm chips (which would be another confusing issue)?
No, because 90% of consumers couldn't give a wet slap whether it is 5nm, 3nm, or a platinum/iridium sponge - and even then, Apple have never suggested that the "M-number" says anything about the process node. It's only confusing for people who got too invested in the rumours that Apple were going to make a chip called the M2 that was absolutely definitely fact triple-exclamation-mark going to use the 3nm process.
Put simply:
M1 (5nm): uses Firestorm/Icestorm cores from the A14 (5nm)
M2 (5nm): uses Avalanche/Blizzard cores from the A15 (5nm) (plus upgraded GPU, neural engine, LPDDR5....)
...so I don't know why people seem to think it somehow doesn't "deserve" to be called M2.
Of course, Apple could have chosen to call it "M1.5", "M1 Bionic", "M100" or "Vera" - because the M-number is a brand, not a specification and Apple can call the next one whatever they like.
Best guess would be that the M2 Pro will be to the M2 exactly as the M1 Pro is to the M1 - more of the same cores, extra I/O, extra RAM channels, same process. However, M2 Pro is also a brand, not a specification Apple could perfectly well decide to use it for something else, as long as the practical upshot was that it was faster than the base M2,
Yes. It does. I wrote ‘effectively end-of-life.’ Apple is no longer supporting my device with its current services. Their solution: buy a new one. Hence, for me and anyone else wishing current support, end of life.
Yet is when? If Apple doesn't start until Q4 22 or Q1 23 the yields will be better. It isn't like there is a magical cut off at TSMC HVM mark that yields don't still continue to get better. At a slower rate but better.
AnandTech Forums: Technology, Hardware, Software, and Deals
Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
www.anandtech.com
"There Yet" as in now , June-August , yes N3 is likely looking somewhat like N7 ( and bit less like N5) at the -Q1 point (or perhaps closer to the -2Q point). The change for N7 from -1Q to +1Q is a pretty decent leap. If Apple were going to stop production after +3Q might be an issue, but higher end M2's on N3 are likely more longer lived than that. So taking lumps for a Quarter isn't that big of a deal. At the markups that Apple is charging for a Mx Max ( > $200 ) even less so.
Pretty good chance N3 isn't going to actually get to HVM until very end of Q3 ( it is now Q2) or very beginning of Q4 . Both of those hit the "broad side of a barn" mark laid down by TSMC ( 2H22). TSMC has said they don't expect to receive revenue for finished N3 product until 2023. So it isn't like HVM is going to start in a couple of weeks from now (i.e., not early or mid Q3; about the first 'half' of that 2H22 is missed. ). The yields in Feburary-March of 2023 will likely look different. By June 2023 even more so.
As for Apple going to eat yields/delays. Apple likely picked N3 years ago. They are along for the ride.
If Apple is using N3 primarily to get a shrink, then the yields will get slightly better with smaller dies. The Pro and Max are exactly 'small' dies. They are not reticle busting but not particularly small. The plain M2 has die size bloat ( like the A15 did over the A14). That is actually bad for the Ultra package because it is limited to 1x recticle size. While one Max isn't at the limit. Two is getting pretty close. It may not have been an option to bloat out the Max configuration die and disrupt the Ultra Package. If have to pull the Max back then pretty easy "trickle down" to a Pro die ( since that is basically a chopped down design of the Max. If have to do Max on N3 doing Pro on N5P doesn't buy much.)
If Apple puts in just 2 E cores ( a full E complex instead of a chopped down one) and adds just two GPU cores to each cluster. Toss in the cache and other uncore updates and likely gets a Pro and Max that is smaller by -15-20% instead of bigger by that same amount over the Pro/Max die sizes. Smaller dies means Apple will eventually get more product from fewer wafers. If they keep the same (or higher markup ... inflation) , then that is more money generated per wafer.
The risk management for to taking on N3 'early' would have been to not try to fill up the same size die size will "more stuff" but to take some substantive shrinkage to get more dies per wafer. Figure out N3 and then go about trying to stuff "much more " into a N3-generation wafer at the same size as the first iteration. ( the old 'tick / tock' model).
If N3 has started HVM at the beginning of Q3 then they probably could have finished off the transition by the end of 2022. They probably aren't. It isn't the end of the world (lots of products over last two years got delayed due to factors outside of company's control). So even if M2 Pro/Max/Extreme slid into 2023 the vast bulk of the Mac line up would have been on M-series. Inertia wise it will not make that much of a difference. The M2 power ( MBA , MBP 13" , Mini , and possible iMac 24" ) are the vast bulk of Mac sales anyway ( which are on a lower risk N5P which there should have been close to zero doubt could use as it over a year old by mid-2022. ) The lower margin stuff is not on N3.
Can I get a tl;dr version?Yet is when? If Apple doesn't start until Q4 22 or Q1 23 the yields will be better. It isn't like there is a magical cut off at TSMC HVM mark that yields don't still continue to get better. At a slower rate but better.
AnandTech Forums: Technology, Hardware, Software, and Deals
Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
"There Yet" as in now , June-August , yes N3 is likely looking somewhat like N7 ( and bit less like N5) at the -Q1 point (or perhaps closer to the -2Q point). The change for N7 from -1Q to +1Q is a pretty decent leap. If Apple were going to stop production after +3Q might be an issue, but higher end M2's on N3 are likely more longer lived than that. So taking lumps for a Quarter isn't that big of a deal. At the markups that Apple is charging for a Mx Max ( > $200 ) even less so.
Pretty good chance N3 isn't going to actually get to HVM until very end of Q3 ( it is now Q2) or very beginning of Q4 . Both of those hit the "broad side of a barn" mark laid down by TSMC ( 2H22). TSMC has said they don't expect to receive revenue for finished N3 product until 2023. So it isn't like HVM is going to start in a couple of weeks from now (i.e., not early or mid Q3; about the first 'half' of that 2H22 is missed. ). The yields in Feburary-March of 2023 will likely look different. By June 2023 even more so.
As for Apple going to eat yields/delays. Apple likely picked N3 years ago. They are along for the ride.
If Apple is using N3 primarily to get a shrink, then the yields will get slightly better with smaller dies. The Pro and Max are exactly 'small' dies. They are not reticle busting but not particularly small. The plain M2 has die size bloat ( like the A15 did over the A14). That is actually bad for the Ultra package because it is limited to 1x recticle size. While one Max isn't at the limit. Two is getting pretty close. It may not have been an option to bloat out the Max configuration die and disrupt the Ultra Package. If have to pull the Max back then pretty easy "trickle down" to a Pro die ( since that is basically a chopped down design of the Max. If have to do Max on N3 doing Pro on N5P doesn't buy much.)
If Apple puts in just 2 E cores ( a full E complex instead of a chopped down one) and adds just two GPU cores to each cluster. Toss in the cache and other uncore updates and likely gets a Pro and Max that is smaller by -15-20% instead of bigger by that same amount over the Pro/Max die sizes. Smaller dies means Apple will eventually get more product from fewer wafers. If they keep the same (or higher markup ... inflation) , then that is more money generated per wafer.
The risk management for to taking on N3 'early' would have been to not try to fill up the same size die size will "more stuff" but to take some substantive shrinkage to get more dies per wafer. Figure out N3 and then go about trying to stuff "much more " into a N3-generation wafer at the same size as the first iteration. ( the old 'tick / tock' model).
If N3 has started HVM at the beginning of Q3 then they probably could have finished off the transition by the end of 2022. They probably aren't. It isn't the end of the world (lots of products over last two years got delayed due to factors outside of company's control). So even if M2 Pro/Max/Extreme slid into 2023 the vast bulk of the Mac line up would have been on M-series. Inertia wise it will not make that much of a difference. The M2 power ( MBA , MBP 13" , Mini , and possible iMac 24" ) are the vast bulk of Mac sales anyway ( which are on a lower risk N5P which there should have been close to zero doubt could use as it over a year old by mid-2022. ) The lower margin stuff is not on N3.
That is not what end of life means. So when the iPhone 4S came out, the iPhone 4 was "effectively end-of-life" because it couldn't use Siri? I don't think so. You are mixing up support and new features.
My prediction: M2 anything won't be 3nm. Apple is saving that for M3.
The upgrade to M3 will be far less impressive if being 3nm isn't anything new. Plus, it gives Apple the ability to sell more M2+ devices without cutting into future sales of M3 chips which will be even better.
Apple's model of: M(x) base, then M(x)+ chips for higher end hardware, repeat, seems to be working quite well. If M(x)+ is too good, M(y) seems sort of meh. M2 as it is is getting close to M1 Pro base levels.
The M2 Pro die design is likely a "chopped down design" of the M2 Max design. The Max die is also a component to the Ultra. The Ultra is pretty much at the 1x reticle limit of TSMC's InFO-LSI packaging technology. So Apple really afford to bloat out the Max die. If that means busting the InFO-LSI limit then loose the Ultra. Which means the small number of GPU and E core updates that the "A15/M2" era are percolating through can't really be an easily tolerated "bloat". Apple needs some space. N3 provides density. Apple doesn't have to change the basic microarchtecture. It is still a M2 generational SoC from a software running on hardware perspective. It is just a smaller dies.
Smaller dies isn't going to hurt. It allows Apple to get more dies out of a single wafer. The wafer costs for N3 are higher. So Apple can defray increase wafer and some interim yield drops by paying for it out of the Pro/Max/Ultra/Extreme markup charges ( which as not small).
The 'plain' M2 can far more easily be decoupled from the Pro/Max/Ultra/Extreme because it has the bulk of the volume. ( a couple of iPads, highest selling Mac (MBA), second highest selling Mac (MBP 13) , Mini , possibly iMac 24" , etc. ). The rest combined aren't even that large in volume. So where the Max goes the Pro die likely also goes. The volume isn't really high enough.
Very similar reason that bulk of die used the "extreme" SoC is also likely to be weaved from a baseline in the Max die. The "Extreme" SoC wlll sell in such relatively low volume it needs another design to help defray development costs with.
( TSMC N3 also could allow an Ultra class single die to be just doubled up (with a large LSI interconnect) for an 'Extreme' if Apple wanted to go that way. If there is enough Studio/Mac Pro / etc volume to support that big of a leap. There are some Pref/Watt wins by making the Ultra one die that Apple may not want to pass and sign up to a large die. Probably some Ultra/Extreme price creep here if they go that way. ). Smaller makes putting more than 4 dies on a single package (not InFO-LSI, but another substantively more expensive option) more tractable.
Right now, it is way too early to call TSMC N3 yield rates bad. It isn't finished yet. Is that simple enough for you?
I am not calling them bad for one main reason, I have no idea what they are. However, given that TSMC has stated the mass volume for N3 isn't till later this year it stands to believe that perhaps the yields are not where they want them to be which is why they are still working on it.
I have a room full of Apple devices dating back to 1980s - macs that power up, but cannot do many useful things. To me, that is end of life; not what some corporate engineer or beancounter declares.
Apple, one of the world’s largest corporations, is increasingly doing underhanded things to customers. Many of us with a third-Gen iPad were told ‘these are powerful devices and you should just wait for the OS to catch-up.’ Now we find out the new OS services won’t even work on our devices. That to me represents end of life.
Having experienced this, and seeing Apple’s vigorous chip updates, I have reasonable doubts about what other ‘powerful’ hardware will be left behind and effectively end-of-lifed. A 3nm die is very advanced. Simply, I don’t trust Apple any longer not to leave me with obsolete equipment that, for me, is end of life.
Last edited by a moderator:
It won't be a huge upgrade; it'll likely be comparable to M2 gains over M1. Nice, but definitely not something that'll render your computer obsolete by any stretch. M1 Pro is gonna be a serious beast for years to come. Plus, once M2 Pro does drop, prices for M1 Pro will come down (esp on the used market) and you can save a pretty penny and still get some serious bang for your buck.
Speaking from experience I think buying the M1 Pro generation will last a long time. Improvements at this point will be incremental until Apple refreshes the design again.
I wouldn’t mind waiting until the end of the year if we can expect to see the M2 Pro chips around October. If it’s going to take until 2023 I might just purchase the M1 Pro this summer.
I wouldn’t mind waiting until the end of the year if we can expect to see the M2 Pro chips around October. If it’s going to take until 2023 I might just purchase the M1 Pro this summer.
You are using an industry wide term inappropriately. Bottom line.
Last edited by a moderator:
I fully expect M2 Pro, M2 Max and M2 Ultra will use the same Avalanche performance CPU cores, Blizzard efficiency CPU cores, A15-class GPU cores and A15-class Neural Engine cores as the base M2 does. It will also use the same Video Encoders/Decoders, the same USB4/TB4 controllers and the same LPDDR-5 RAM (though with a higher-speed memory controller for allow greater bandwidth).
So they are no more different chips with the same name than the M1 is compared to the M1 Pro/Max/Ultra.
Yes, the fabrication process node will be different (should Apple be able to move to TSMC's 3nm process) and that will allow Apple to clock the cores higher and/or have them consume less power, but there is no technical reason that would prevent Apple from moving the base M2 to the same 3nm process and doing the same.
If M2 Pro is coming to Mac mini next year perhaps it'll be binned versions due to as yet unknown yield rates for M2 Pro? That's my first take on the situation as minimum package will surely be 16Gb with 512Gb storage - making a high end Mini easily approach $1499.
And the existing case would then surely be the shoo-in to help power and cool the arrangement while keeping costs down.