Yes, that's it. I think the core count is pretty stable at 2P+4E. Cache is an easy target for making the die size smaller.
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.
Where are those 3nm Apple Silicon products?
- Thread starter hajime
- Start date
- Sort by reaction score
Yes, that's it. I think the core count is pretty stable at 2P+4E. Cache is an easy target for making the die size smaller.
theres been a slowdown in Mac sales the last few quarters.
and contractually, apple must purchase x wafers from tsmc at each node.
so until that contract is fulfilled, they are not going to release the m3 Macs.
the same thing with ram chips. until those ram contracts are filled up and new ones are negotiated, apple isn't going to add more ram into the iPhone.
and contractually, apple must purchase x wafers from tsmc at each node.
so until that contract is fulfilled, they are not going to release the m3 Macs.
the same thing with ram chips. until those ram contracts are filled up and new ones are negotiated, apple isn't going to add more ram into the iPhone.
Just looking for clarification here:
Did you mean apple has an existing contractual obligation to purchase wafers with M2 SoC chips from TSMC?
Utilization for TSMC N5 is nearly at full capacity. Everybody is buying GPUs for AI applications. TSMC can't even fill that demand until 2025. Nvidia H100 cards are selling for $50k on eBay. As Elon Musk puts it, "GPUs at this point are considerably harder to get than drugs." Microsoft and AMD are also developing their own AI chips which will use N5.
Apple could sell their N5 wafer slots in a second. I seriously doubt Apple was caught off guard or would let a slowdown in sales dictate their tech roadmap. Slower Mac and iPad sales was predicted quite a while ago. Consumers aren't buying M2 if they have a perfectly working M1, especially if people have returned to work and school.
The base N3B process used for A17 Pro began MP last year. But the simplified N3E process for M-series chips isn't ready today. It only began MP in mid-2023.
Is crypto mining still a big consumer of high end GPU too?
yes, thats usually how it works. because tmsc spends billions developing and setting up the manufacturing lines, which they dont ask their customers to pay for,
but in return, they force you to contractually buy x wafers at x time for x length.
its a hard contract where you can be penalized (fined) for not taking delivery of those chips.
so I presume because Mac sales have slowed, that there may be a surplus of m2 chips that apple still is contractually forced to purchase.
im sure on the internet its really easy to say
"apple can just sell their capacity back"
the truth is, no one knows for sure because
1/ apple has preferential treatment at tsmc. this could mean specific clauses in their contracts which are beneficial only to apple or beneficial only to tsmc. it would be stupid to speculate that
"well they can just give it back"
no one but probably 40 people in the world know the exact exact details of that.
what I wrote above is the "general" rules that firms like amd/nvidia/etc abide by when they purchase wafers thru tsmc
TSMC is a contract foundry. It's their job to shuffle customers around. At the end of the day, it's about money. Nvidia chips are in such high demand right now and they have the money.
Apple could juice demand by putting M2 into iMac and iPad Air, dropping the price of M2 MBA, and adding cheaper variants of MBA 15. The fact Apple hasn't done any of that suggests they aren't concerned about demand.
Most likely A17 standard is based on design rules of A16 which is supposedly on 3nm node but couldn't make it on time. Plus A17 going to feature not only in future iPhone 16 and low cost iPhone SE; not on iPhone 15 Pro/Max anyone. A17 Pro seems only available for one year...
It is the future of N3B I am interested; will Apple going to maintain two process nodes in parallel? Let N3B handles A18 Pro going forward? M3-based PCs are coming soon, so it is likely based on N3B, how about future M3 Pro/Max and M4 going forward???
Last edited:
A16 couldn't have been designed for 3nm because the N3 schedule (volume start in 2H'22) always meant it was for 2023 iPhones. Given the wafer cycle time is four months, it means Apple wouldn't have designed A16 for N3. Same thing for M3 chips - those use N3E so they won't launch until very late 2023 or early 2024.
I think there's zero chance Apple will continue using N3B for A18 Pro given what we've heard about yields. If the yields are poor with 19 billion transistors, how can it handle A18 Pro? It makes even less sense to use it for M-series chips with 2X, 3.5X more transistors.
The rumors spun that as a chase , but A16 on TSMC N3 never lined up with any TSMC released timelines at all.
In mid 2019 TSMC had not even fully defined N3.
TSMC: 3nm EUV Development Progress Going Well, Early Customers Engaged
So hard to believe that Apple was 'betting the farm' on 1H 2023 release of a fab process that wasn't even fully flushed out in 2H 2019. By 1H 2020 Apple and TSMC were in world wide pandemic. So 1H 2023 would be even LESS creditable.
By August 2020
TSMC was placing N3 in 2H22 ... which would be too late for A16 ( which would need to start production in March-May even on older, faster N5 family ... let alone the > 1 month longer to make N3. More like Feb-March ).
The notion that Apple suddenly at the last minute have to divert A16 to N4 doesn't add up. N3 never was a viable candidate for A16 timeframe requirements for any decent length of time. ( there is a narrow timeframe in Sept '19 - March '20 when someone could have been in denial ... but after that. ). More likely Apple 'wished' N3 happened to be early and it wasn't so some stuff had to wait for a good window.
There was lots of extra stuff to push into the A-series dies and they were bloating up anyway. So the 'even more stuff' had to wait.
M4 ... pretty decent chance those are N3P or N3S.
" ... N3's improvements do not stop with N3E. TSMC is set to bring out N3P, a performance-enhanced version of its fabrication process, as well as N3S, density-enhancing flavor of this node, some time around 2024. Unfortunately, TSMC is not currently disclosing what improvements these variants will offer compared to baseline N3. ... "
" .. Since N3P is an optical shrink of N3E, it will preserve N3E's design rules, enabling chip designers to quickly reuse N3E IP on the new node. .. "
If Apple releases M3 in Q4 2023 then about 18 months after that is in 2025 which would line up with H2 2024 for N3P.
There is no good reason for Apple to hand onto N3B too far past the end of 2024 - mid 2025 (except maybe as 'hand me down' M3's into iPad Airs).
N3B has major problems with Apple being the only customer because Apple is likely also to move on from N3B ( and N3E) . At least N3E has a design compatible 'off ramp' ( N3P , N3S). [ Some folks hand wave at N3X , but Apple won't touch that with a 50ft pole. It throws Perf/Watt out the window. Apple doesn't want that. ]
By 2025 or so there should be second gen , high HA, EUV fab machines that make N3S ( and maybe N3P ) just way better options for what Apple has typically done with their high footprint cache designs. N3B -> N3P/N3S makes more sense if want to avoid the entire N3E backslide on density for M-series.
Once Apple takes high volume dies off of N3B it just probably isn't a viable long term process for TSMC. The process sucks up more than its fair share of EUV machines ( too much multipatterning time when wafer demand from rest of N3-family lineup is quite high. Same machines applied to those would get higher wafer throughput rate. ).
The A18 Pro ... probably is on N3E. Just a bigger die with some incremental improvements and deeper N3-family optimizations. The A17 possibly ported to N3E for multiple year production via a series of 'hand me down' products as a 'new' release. If A17 isn't ported to N3E then it is all the more likely that 'plain' A18 will need to be one N3E ( that will 'pulled' the A18 Pro there. Very likely just either a relatively small die trim or a binning that is difference between A18 and A18 Pro. Same baseline design costs distributed over the aggregate long term volume. )
The M3 squats on N3B because most of the line up is a one-shot packages that have no 'Hand me down' transition products to go to M3 Pro , Max , Ultra. The plain Mn long term hand me down of iPad Air can do a combo of squatting on a stockpile of M3 and perhaps no quite so delayed transition to M4. Once most of the volume of iPhone and Macs get off of N3B ... it is likely a 'dead' process. Mainly, a side effect of Apple being just about the only customer.
[ NOTE: if Apple chose to pragmatically delay M3 until 2024 to put it on N3E then N3B is even more 'dead' quicker. THe aggregate Mac volume could keep N3B going longer but no singular Apple SoC package is going to keep it alive. ( not A18 Pro or M4 would work by themselves over a short period. ) ]
Last edited:
They can keep selling M2 devices for years, such as introducing an M2 Macbook SE.
No. A 32nm VS (45NM original) improved A5 chip was released mid life for the iPad 2 without press/mention.
Look into the far past and the 3GS Samsung processor also underwent a slight hardware change mid-life to negate a hardware jailbreak (old vs new bootrom.) No performance difference there however.
Testing shows 32nm A5 chip improves iPad 2 battery life by as much as 16% | AppleInsider
A series of tests performed on the latest version of Apple's iPad 2 show that the move to a 32nm-process A5 chip results in battery life improvements of as much as 15.8 percent.
appleinsider.com
Last edited:
If my high school was any indication, drugs aren’t particularly hard to get.
Still, this isn’t the first braindead statement from Musk.
He's just talking to his fan base who are likely drug users.
And there is no reason for Apple customers to really care about any internal variations if the capabilities and price of a product remain the same. What goes on under the hood is irrelevant to most people. So Apple could switch an M3 using product from an M3 made on one kind of TSMC process to another and whatever performance differences will be so small as to not note.
There has been a slowdown in all PC sales in the last few quarters as we get past the bubble of people buying new computers to start working from home. There were also a lot of rumors of a recession over the past year that likely caused people to defer purchases. Now that the economy seems more stable again, maybe some of those will come back to the market.