I always assumed that Apple gave their products the Air branding because it is very light, not because it is fan-less tho.
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.
Could we see M3 before A17?
- Thread starter senttoschool
- Start date
- Sort by reaction score
That was the original reasoning, but since the switch to AS, the fanless and silent operation of the MBA has become a talking point as well.
Three important things to note:
1. At last report, TSMC was only getting around 45,000 wafers/month from the newer process. With the 5nm process used by M2, they have been getting around 195,000 wafers/month. Fewer wafers means fewer finished SoCs, even if the M3 is on a smaller die.
2. Because not every SoC on the wafer will pass testing/quality control, that yield of finished product drops more. While it is usually a very small percentage of the overall production, this is a new node and yields tend to be lower in the beginning, improving as the production processes become more refined.
3. Depending on how testing goes, these new SoCs could be binned based on number of viable GPU cores, like we have seen in the base M2 MacBook Air. (i.e. a config with an 8 core GPU and CPU and a config with a 10 core GPU). There are other factors at play, including whether they are manufacturing just the base M3 or variants such as an M3 Pro/Max/Ultra.
Chuckle. Funny how it already works on the OS foundation that Apple has already worked on and provided.
Echo III Desktop - Tech Specs - SONNETTECH
View videos and tech specs for Echo III Desktop, determine hardware and OS compatibility, and see recommended accessories.
Go to the Supported PCI-e cards PDF file there and will find
Blackmagic Declink 8K Pro with a 'Yes' in the supported for M1/M2 column but with a footnote number 8. That footnote states : " that x8 PCI-e v3 may not achieve maximum bandwidth, or number of video channels..." due to limitations of x4 PCI-e limits of TBv3 (and v4).
Apple cared enough that it is mostly working now. The whole "apple doesn't care" doesn't really have a good foundation. What proposing here is that Apple purposely kneecap the potential for what the software already can do. That is more than silly. Perhaps folks are drinking that much kool-aid in Cupertino so that looks 'sane'. Not sure why would want to continue buy long term from folks with that kind of mindset.
The 'missing part' is the 'better than Thunderbolt' bandwidth. It is hardware, not software, that is the bottleneck.
It is mainly a bozo move for the Mac Pro not to uncork that bottleneck for which software and the PCI-e cards are already there for macOS. It is a decent internal PCI-e controller that is missing. That could fit is very well orthogonally with the rest of Apple's basic M-series building blocks used in the rest of the line.
x8 ( or x16) PCI-e v4 cards for storage , networking , etc ... Same problem. Software is there but hardware "leaning too hard on Thunderbolt" is a major stumbling block. Apple mentioned "leaning too hard on TB" as a stumbling block back in 2017 . It would baffling as to why they would not drag the same conclusion when PCI-e slots moved on to v4 and v5 and TB was still on v3. There are demos of x4 PCI-e v5 SSD being shown in 2023. "v3 is good enough in 2023-2027" is a joke.
Either flip 4 TB controllers on die into being a x16 PCI-e v4 block ( half two larger building blocks that differ slight on added I/O profile as combine them. ) . Or add in a couple of PCI-e controllers on a chiplet ( and could keep the excessive number of TB controllers... only obscure corner cases need more than 6. 8 is basically ridiculous and > 8 is just plain 'wrong' and clearly wasteful. Neither one of those requies any signficant changes to other cores or memory subsystem.
I don't think Apple has to necessarily chase after 3rd party display GPUs, but limiting themselves to x4 PCI-e v3 for a Mac Pro in 2022-23 is 'nuts'. Even more so if the Mac Studio Ultra is going to market against a Mac Pro Ultra. the Mac Pro better have some production differentiated added value. x16 PCI-e v4 (maybe eventually v5) versus x4 that is one (or two) generations back is a pretty good gap. Two of those an even bigger gap.
Apple could do some Rube Goldberg kluge with using "extra" TB controllers to provision internal slots. That would be a joke.
I doubt the new Mac Pro will be an ultimate temple to hyper modularity. Likely the bulk of the internal archecture will be in upgraded building blocks used for the other M-series packages. Same cores, memory stack , etc. ,but with a narrow set of I/O augments. It won't be an everything for everybody system (e.g, not a Nividia 4090 'killer' ), but cover a decent size class of workloads that the rest of the line up has trouble with.
With the 3nm process, Apple is the only customer...
With the 5nm process, Apple was not the only customer...
This is silly. There are well over a dozen other N5-family customers at TSMC. TSMC doesn't only do stuff for Apple only.
45K/month is relatively a very large bucketload when there is only one customer. The rumors that put the capacity at 45K/months also said that the actually capacity utlization was only 50%. So half the capacity is sitting around idle. Apple isn't using it either at the moment. 35K, 45K , 55K, 65K , or 155K capacity ... doesn't really matter if all that is coming out is 22K/month.
If the M3 were about the size of the M1 die 45K capacity is far over the limit needed to do all the dies Apple would need outside of an initial demand bubble. Demand bubbles can be 'smoothed out' by just stockpiling longer in advance. To do 1M M1 dies per month is less than 3K wafers per month. If do that for 12 months that is 12M dies. total number of Macs sold in year is in the 25-30M range. 20K wafers a month would still be overkill.
An A15 size die 10M dies/month is 23-25K wafers per month. 45K swamps that also.
Beside the initial Sept-October iPhone demand surge. 45K is very decent amount for what Apple needs with the restrictive use of the A17 only in the iPhone Pro and likely staged releases of M3-family SoCs. Apple can't do some kind of 'big bag' release where update entire product lines of iPhones , iPads, and Macs all on the lastest node.... but does Apple really need to do that anyway. Most of the time Apple is shipping older than newest bleeding edge SoCs in most systems in the product line.
The numbers above are for 0.2 defects density rate. There are dead dies in there. (edit. should have said account for in there. the 1M , 10M dies for month above are for 'good' dies that don't have any defects at all.) N3 has already largely already done that. It got to HVM status back in December ( after being in high risk status most of the year of 2022. There were many , many months of walking down done). HVM status means the defect status is 'tolerably low enough' in TSMC's eyes. It is more costly wafer so maybe Apple is a tad more skittish, but it isn't going to get any better doing nothing. Production utilization rates according to that rumore were much lower than 50% in the first couple of months. If true it likely has been a ramp. ( the 3 months of stockpiling is probably overblown as a connotation. If went 25 , 30 , 35 , 45 , 50 % utilization ramp then have been walking the ladder on actual production rates. 25% of 45K is ~11K which is much different than ~22K. 3 months at 22K would be a much bigger pile.
AT 50% production capacity utilization rates ... having the necessary number of dies to bin probably is not a huge problem.
Binning is way , way , way overblown. The bulk of the stuff that sold binned is likely good cores turned off. If price the 'binned die' end user cost to cover the number of good die costs for the whole wafer, then the binning stuff is mainly to pad out the product margins. Not to struggle to get enough working product.
Defects that land in an internal bus node , memory controller , display controller , etc go into the trash. Binning on solely CPU or GPU cores isn't controlling for all of the defects. It is just picking up incremental extras 'for free' at the margins.
Last edited:
Idk, seems like you kind of proved my point. Maybe I'm wrong but it kind of seems like it works on accident. They didn't care, otherwise there wouldn't be crippling bandwidth issues using those cards or other cards that do other things Apple doesn't really care about.Chuckle. Funny how it already works on the OS foundation that Apple has already worked on and provided.
Echo III Desktop - Tech Specs - SONNETTECH
View videos and tech specs for Echo III Desktop, determine hardware and OS compatibility, and see recommended accessories.www.sonnettech.com
Go to the Supported PCI-e cards PDF file there and will find
Blackmagic Declink 8K Pro with a 'Yes' in the supported for M1/M2 column but with a footnote number 8. That footnote states : " that x8 PCI-e v3 may not achieve maximum bandwidth, or number of video channels..." due to limitations of x4 PCI-e limits of TBv3 (and v4).
Are you really suggesting they are building custom drivers on their new CPU architecture “by accident”? That’s some deliberate accident 😁
And regarding the bandwidth limitation, well, that’s the limitation of current external I/O. It’s not like many other laptops can do better.
Last edited:
Reality check and back on en envelope calculations: 45000 wafer per months. These are the big production wafers and not the small 2 and 4 inch I use in research so assume 50-100 Ultra chips per wafer* = 2 250 000- 4 500 000/month*6 months=13.5-27 million M3 Ultras in stockpile. If Apple is going to sell >10 000 000 ultras is MP and studio in a short time scale to warrant the stockpile, the M3 Ultra really needs to have large boost in performance/price ratio. Smells more like the M3 for 15 inch Macbook Air.
*I know fully well that Ultra= 2 Max but we talk about the MP here to counting Ultra make sense. I assume 40 000 mm2 is useful are for a 300 mm diameter wafer and the M3 Max is about 400 mm2.
One 300mm wafer could produce around ~350 M3 chips assuming that the size will be similar to M2.
45,000 wafers * 350 chips per wafer = 15.7 million M3 chips per month. Apple sells around 7 million Macs per quarter. Hence, Apple only needs about 2 weeks of 3nm production to fulfill an entire quarter of M3 Mac demands.
This begs the question: What the hell is Apple making on 3nm since December 2022? There are reports that Apple is the only 3nm customer at the moment.
Edit: 4 months * 45,000 wafers = 180k wafers * 350 chips per wafer = 63 million M3 chips produced between December and now. Again, Apple sells ~7 million Macs per quarter.
If Apple is all-in on M3, we could see all M products get updated in quick succession. This means we could see enough M3 supply to update the 13" Air, launch 15" Air, iPad Pro 13", iPad Pro 11", iMac, Mac Mini all get updated with M3 relatively quick. We could also see M3 Pro/Max/Ultra chips come out sooner than expected.
This calculator suggests 376 M3 chips per wafer. Let's assume that defects knock out 26 chips. That's still 350/wafer.
Last edited:
Think your calculator need new batteries. Haha.
no doubt about that...IF all the rumours about TSMC are true...maybe mass production just started this or last month,and if its true that Apple is the only customer for these chips..i dont know
yea, senttoschool needs new school
I didn't have my morning coffee. 😅
4 months * 45,000 wafers = 180k wafers * 350 chips per wafer = 63 million M3 chips produced between December and now. Again, Apple sells ~7 million Macs per quarter.
Point stands. What hell has Apple been doing with all the 3nm wafers?
Again, we are not sure if TSMC is building since late December or late March...is not like the CEO told us that...also is really Apple the only customer for the N3? And third, maybe Apple now likes to have in stock the chips and rely on those for upcoming macs and ipads and iphones instead of building the devices without having enough chips...i really dont know but there are still too many questions and we all know Apple is very careful when and where money are involved...
But im glad if there are enough chips and no more "supply chain delays" because of that
Who can say for sure that TSMC only produces N3-based chips?
www.anandtech.com
TSMC's 3nm Journey: Slow Ramp, Huge Investments, Big Future
PR released on December 29th 2022:
As for if Apple is the only customer for 3nm right now, this seems to be true. No one else has publicly acknowledged a 3nm chip. If Apple isn't the only customer, surely we would have another company come out and boast about how they have the world's first 3nm chip.
I'm guessing that Apple is indeed hoarding 3nm M3 chips and that we will see a lot of Macs and iPads get updated to M3 after WWDC.
Those numbers are also assuming that all of the SoCs being produced are the base M3. If they are also manufacturing M3 Pro and M3 Max, then that chip yield drops by a noticeable amount. Furthermore, given that every iPad currently sold by Apple (excluding the 9th and 10th gen base iPads) uses M-series SoCs instead of the A series, a lot of those chips could never even see their way inside of a Mac laptop or desktop.
reconciling! (for my own benefit)
3,000 wafers * 333 chips/wafer = 1,000,000M M3 dies
25,000 wafers * 400 chips/wafer = 10,000,000 A15 dies
Plenty of capacity
Also - Plenty of capacity
333 chips/wafer vs 350. Both high-quality educated guesses, and both noting that TSMC has plenty of room for A17 and M3 production in 2023 with their 45,000 wafer/month (and growing) capacity.
And therefore, most importantly:
M3 hardware announcement at WWDC
Or:
3,000 wafers * 333 chips/wafer = 1,000,000M M3 dies
25,000 wafers * 400 chips/wafer = 10,000,000 A15 dies
Plenty of capacity
Also - Plenty of capacity
333 chips/wafer vs 350. Both high-quality educated guesses, and both noting that TSMC has plenty of room for A17 and M3 production in 2023 with their 45,000 wafer/month (and growing) capacity.
And therefore, most importantly:
M3 hardware announcement at WWDC
As i said, its almost confirmed
"Gurman said the M3 chip is "coming later," suggesting that the initial 15-inch MacBook Air model will be powered by the M2 chip or a very similar chip."
So 15" Macbook air is coming probably by WWDC with M2
So the N3 will come in the fall...probably A17 first and after the ipads and macs
"Gurman said the M3 chip is "coming later," suggesting that the initial 15-inch MacBook Air model will be powered by the M2 chip or a very similar chip."
So 15" Macbook air is coming probably by WWDC with M2
So the N3 will come in the fall...probably A17 first and after the ipads and macs
15-Inch MacBook Air With M2-Like Chip Spotted in Developer Logs
An unreleased 15-inch MacBook Air with a processor "on par" with the M2 chip has been spotted in App Store developer logs, according to...
www.macrumors.com
I highly doubt it.
You can easily calculate it yourself if you know the dimensions of the SoC.
The yield % wouldn't drop by much because Apple has built-in ways to cut out a defective CPU/GPU core.