OK.
There's no obvious tech reason the AVX2 support should work only on M4, and I did not even see such a claim until your post. Seems like there's an industrial strength level of cluelessness surrounding this issue...
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.
M3 vs M4
- Thread starter Dulcimer
- Start date
-
- Tags
- m3 chip m4 chip neural engine
- Sort by reaction score
There's no obvious tech reason the AVX2 support should work only on M4, and I did not even see such a claim until your post. Seems like there's an industrial strength level of cluelessness surrounding this issue...
It's just from the one guy as far as I know, but unfortunately he is a popular YouTuber with a decently large audience on Twitter as well and he makes these kinds of errors both in posts and videos frequently. @Xiao_Xi had linked to the tweet in question but this time the YouTuber deleted the offending tweet sometime yesterday so it no longer shows up.
Last edited:
I assume so. Everything reported about N3B makes it seem like something Apple would want to get off of quickly. So M3/A17 products are gone once N3E replacements show up and no “cheaper” devices with those chips and M2/A16 will be used for those instead.
I don't think Apple ever kept the previous generation iPhone Pro models around - e.g. The iPhone 14 Pro was discontinued a year after its release and so was the iPhone 13 Pro. There weren't any standard non-pro A17 iPhones, so the discontinuation of the A17Pro doesn't really tell us much. As such, given the products released, the only M3/A17Pro product Apple was ever likely to keep was the M3 Air. Could they skip it and keep the M2 Air as the base previous generation model? I guess. But that seems unlikely.
Also keep in mind that everything is relative. N3B may be troubled by TSMC standards but not only is Intel producing LNL on it, but Intel announced that they have canceled Intel produced A20 Arrow Lake and will likely be producing it on TSMC N3B instead.
That is a bit backwards. In both cases it is more so "lack of customers using the node" that is the 'problem' more so than an technical issue with the node. Long running leaks of Arrow Lake was always on N3B.
2021
2022
Intel Arrow Lake-P with 320EU GPU confirmed by a leaked roadmap, targeting to compete with Apple 14" premium laptops - VideoCardz.com
Intel Arrow Lake-P GPU confirmed feature up to 320EUs Just a few days ago Intel officially confirmed it will launch Arrow Lake architecture by 2024. This successor to Meteor Lake is now confirmed to utilize a combination of three process nodes. It will also take advantage of Intel’s new die...
Arrow Lake at one point was suppose to have a more laptop focused variant, but that too probably got the budget chop to chase 20A. And now they likely have neither. ( likely a downclocked desktop variant will target laptops with dGPUs. Like previous generations.)
"... In the case of Core Ultra 5 240F, a change of CPU tiles may also mean a change of fabrication node. The 6+8 version is said to be using Intel 20A, while the 8+16 is rumored to use either Intel 20A or TSMC 3 nm node. ..."
Intel Core Ultra 5 240F "Arrow Lake-S" CPU to use either 8+16 or 6+8 core tiles - VideoCardz.com
Core Ultra 5 240F, expected to use two versions of the Arrow Lake-S silicon A new rumor suggests Intel will continue its ‘tradition’ to mix silicon dies within the mid-range CPU segment. The next-gen Arrow Lake-S series is on track to debut on the desktop platform this year, introducing the...
There always was an Arrow Lake die variant that was only on TSMC. In the higher volume subset of the Arrow Lake target market Intel was going to try to squeeze in 20A to cover some of the volume. (incrementally reducing Intel's consumption of TSMC wafers). Why is Intel throwing more money at making a duplicative die on a different process was always the question.
It is Intel 20A that was a 'late addition' add over time. Adding 20A to Arrow Lake was largely a way of throwing extra money at the product. (duplicate dies). Probably in part to help sell customers of Fab foundary services on whether 20A was something they could buy. In that sense, similar 'problem' that N3B has in that almost nobody wants to buy it. Only Apple and Intel bought into N3B (costs ).
Lunar Lake has Lion Cove cores and so does Arrow Lake. Lunar Lake is also more strategic ( laptops which is most of their consumer CPU market). So there always was going to be N3B cores... it was more so whether some Intel fab process might join them or not.
Intel also shifted their N3B wafer starts by a substantial amount of time. So they likely have a fixed amount of N3B wafers that they "have to" buy. If the project volume of LNL/ARL SoC sales is somewhat down that would be there was even less "extra space" to slot in 20A dies to consume. If the pond for 20A product is getting smaller and smaller, then at some point it doesn't make sense to offer the process. Time and effort is better spent on processes that do have relatively high volume customers.
18A has some customers considering it. Getting the 'bugs' out of that one will make a difference to the Fab business. Too slow helping those customers get to market and some of them will 'walk away'. That is a worse outcome than some largely 'ego' thing of slapping duplicative Intel process on Arrow Lake.
Additionally, Intel has relatively limited EUV production capacity. If the Intel 3 server stuff soaks up most of the short term production capacity and the potential 18A customers would soak up even more in 2025 ... would Intel even have resources for a 'side project 20A' runs that took substantive volume away from TSMC? Perhaps not. One reason Intel is over at TSMC for N3B is not their 'process node' deficit. It is because that can't do very high volume production of EUV of any process. ( AMD has been taking server market share away from Intel. The last problem they need now is to turn away some server customers they have left because the Intel 3 server dies are running constrained on availability some desktop only thing is using capacity they need for the short term. ). Intel has to be very careful of not substantively oversubscribing they limited set of EUV systems.
N3B is going to get canceled too. Once Apple and Intel evolve their product mix to "newer stuff" there aren't going to be any customers arriving as slow adopters to the process. ( it is 'dead end' compatibility kit to other parts of N3 family. )
Apple is still offering the M2 MBA while they sell the M3 one. They did the same thing for M1 when moved MBA to M2.
First, indications are that Apple isn't going to drop the M3 MBA until 2025 ( Q1-Q2).
Is the M3 going to get 'hand me down' to the iPads? Probably not. The Pro is sitting on M2 Air / M4 Pro now. When Pro jumps to M(>4) the Air likely would jump to M4.
Could Apple leave the MBA at M2 and have a MBA M4 + MBA M2 line up? Maybe. Competition wise in the Windows market nobody is hitting the 'snooze' button in the $700-999 laptop space right now. Qualcomm vs Intel vs AMD all are going to be even more contested in 2025. M2 and 8GB is going to 'smell' even more 'stale' if the M4's step up to 16GB minimum. However, Apple will squeeze better margins out of it because it is 'even older stuff'. (wouldn't be first time Apple has a comatose laptop configuration in the line up.)
IF the 'old' MBA moves to 16GB base configuration then the 'even cheaper' M2 could help offset the increase in BOM costs for Apple. Not that the M3 would be problematical... it would just cost more than Apple wants to pay. I wouldn't bet the farm they will move it up though. Probably will shoot for a "good enough for some people" , cheaper costs low end MBA. That will also help herd more folks into buying the M4 variants.
I think it's fairly likely Apple will drop all M3s when M4 Pro/Max ship. The M2 is still a very competitive chip; they could sell MBA M2s with only 16GB and 24GB options, which wouldn't smell stale at all.
This is phrased as though it disagrees with me, but it pretty much exactly what I wrote, albeit greatly expanded - but I let the article I linked to cover the details that Arrow Lake was originally going to be a mix of N3B and 20A but the 20A portion was canceled ... The point was not that 20A was a bad node performance wise or even yield wise, we simply don't know that*, but that Intel, for the reasons you delineated so throughly, still found it more economical to produce on it than their own foundry nodes and indeed have potentially expanded their production on it. Thus there is a limit to how bad the economics of producing N3B must be. In other words the economics of N3B is bad relative to TSMC's other nodes but the economics of the Intel nodes are worse which is why Intel is using it. Perhaps I should've been more clear and detailed, but I thought it was already clear from what I wrote and the article I linked to.That is a bit backwards. In both cases it is more so "lack of customers using the node" that is the 'problem' more so than an technical issue with the node. Long running leaks of Arrow Lake was always on N3B.
2021
2022
Intel Arrow Lake-P with 320EU GPU confirmed by a leaked roadmap, targeting to compete with Apple 14" premium laptops - VideoCardz.com
Intel Arrow Lake-P GPU confirmed feature up to 320EUs Just a few days ago Intel officially confirmed it will launch Arrow Lake architecture by 2024. This successor to Meteor Lake is now confirmed to utilize a combination of three process nodes. It will also take advantage of Intel’s new die...videocardz.com
Arrow Lake at one point was suppose to have a more laptop focused variant, but that too probably got the budget chop to chase 20A. And now they likely have neither. ( likely a downclocked desktop variant will target laptops with dGPUs. Like previous generations.)
"... In the case of Core Ultra 5 240F, a change of CPU tiles may also mean a change of fabrication node. The 6+8 version is said to be using Intel 20A, while the 8+16 is rumored to use either Intel 20A or TSMC 3 nm node. ..."
Intel Core Ultra 5 240F "Arrow Lake-S" CPU to use either 8+16 or 6+8 core tiles - VideoCardz.com
Core Ultra 5 240F, expected to use two versions of the Arrow Lake-S silicon A new rumor suggests Intel will continue its ‘tradition’ to mix silicon dies within the mid-range CPU segment. The next-gen Arrow Lake-S series is on track to debut on the desktop platform this year, introducing the...videocardz.com
There always was an Arrow Lake die variant that was only on TSMC. In the higher volume subset of the Arrow Lake target market Intel was going to try to squeeze in 20A to cover some of the volume. (incrementally reducing Intel's consumption of TSMC wafers). Why is Intel throwing more money at making a duplicative die on a different process was always the question.
It is Intel 20A that was a 'late addition' add over time. Adding 20A to Arrow Lake was largely a way of throwing extra money at the product. (duplicate dies). Probably in part to help sell customers of Fab foundary services on whether 20A was something they could buy. In that sense, similar 'problem' that N3B has in that almost nobody wants to buy it. Only Apple and Intel bought into N3B (costs ).
Lunar Lake has Lion Cove cores and so does Arrow Lake. Lunar Lake is also more strategic ( laptops which is most of their consumer CPU market). So there always was going to be N3B cores... it was more so whether some Intel fab process might join them or not.
Intel also shifted their N3B wafer starts by a substantial amount of time. So they likely have a fixed amount of N3B wafers that they "have to" buy. If the project volume of LNL/ARL SoC sales is somewhat down that would be there was even less "extra space" to slot in 20A dies to consume. If the pond for 20A product is getting smaller and smaller, then at some point it doesn't make sense to offer the process. Time and effort is better spent on processes that do have relatively high volume customers.
18A has some customers considering it. Getting the 'bugs' out of that one will make a difference to the Fab business. Too slow helping those customers get to market and some of them will 'walk away'. That is a worse outcome than some largely 'ego' thing of slapping duplicative Intel process on Arrow Lake.
Additionally, Intel has relatively limited EUV production capacity. If the Intel 3 server stuff soaks up most of the short term production capacity and the potential 18A customers would soak up even more in 2025 ... would Intel even have resources for a 'side project 20A' runs that took substantive volume away from TSMC? Perhaps not. One reason Intel is over at TSMC for N3B is not their 'process node' deficit. It is because that can't do very high volume production of EUV of any process. ( AMD has been taking server market share away from Intel. The last problem they need now is to turn away some server customers they have left because the Intel 3 server dies are running constrained on availability some desktop only thing is using capacity they need for the short term. ). Intel has to be very careful of not substantively oversubscribing they limited set of EUV systems.
*Reuters did suggest that Broadcom was unhappy with Intel's progress on 20A's sister 18A node, but that's not confirmed by official sources and the reason wasn't exactly clear - possibly, maybe yield. If so, it's entirely possible that given Intel's tight Node development 20A wasn't going to be any better yield than N3B for the volume launch this year on top of all the other issues described above.
Yes of course N3B will one day get canceled too, the question is when? Personally I think if Apple found the economics of producing M3 on N3B undesirable, they wouldn't have added it to the Air at all, especially the 13" Air - just as they didn't add the M3 to the iPad Air and chose the M2 instead. Of course, volume might've limited the M4 coming to the Air right away but we know the M4 has been ready for a while.
Possible. But the M3 Airs just launched in the Spring - just in March in fact, if the economics of the M3 were that bad and were going to be replaced in the fall by November, why bother releasing the Spring version at all? A product run of 8 months is not unheard of in Apple land but it isn't common either. Further, that would break the pattern of selling the base RAM models of M1 and M2 Airs when the subsequent generation hit, but of course you're already suggesting that Apple will break that pattern by not selling the M3 at all - substituting a processor bump for a RAM bump. Certainly possible, but I think less likely than simply continuing the pattern so far established of just selling the M3 in the base Air when the M4 launches in the Air, presumably next spring.
Last edited:
My expectation is a 2025 Q2 M4 MacBook Air release, in time for the Back To School 2025 promotion.
If it comes with 12-16 GB RAM and 256 GB storage base, I will buy a base model for my wife, to replace her 2017 MacBook Air with dual-core Core i5, 8 GB RAM, and 256 GB storage. Actually, the 8 GB RAM is fine for her, and even the speed is mostly fine for her too, but the lack of hardware h.265 HEVC decode support has become a problem in the last year or so. But I figure if I'm going to get a new machine, I may as well get one with more RAM since we keep our machines a very long time.
2025 Q1 would work better for us since my wife's birthday is in Q1, but I don't think that will happen. And it won't happen in 2024.
BTW, I wouldn't be shocked if Apple just discontinues the M3 MacBook Air when they release the M4 model, and continues to sell the M2 model as the budget entry-level model. I get the impression that TSMC is keen to kill off N3B manufacturing sooner rather than later. However, I wouldn't be surprised either if Apple kept M3 around. I'm kinda 60/40 on the M2/M3 likelihood when M4 is released.