"Total operating expenses decreased by TWD2.6 billion, mainly as 5-nanometer technology moved from R&D stage to mass production during the first quarter."Maybe I missed it, but AFAIK just about nobody was claiming that TSMC was in high volume mass N5 production in Q1 2020, not even TSMC themselves.
While most of your post absolutely makes sense (Apple controlling OS and main system APIs leading to highly optimized hardware), this last sentence is a little disappointing.People are underestimating how much success Apple will have with ARM Macs.
I completely disagree with you.While most of your post absolutely makes sense (Apple controlling OS and main system APIs leading to highly optimized hardware), this last sentence is a little disappointing.
ARM Macs will be great as another Apple-ecosystem device. Which means they will cannibalize iPads.
For example: ARM-powered Macbook Air will essentially be an iPad Pro with a fixed keyboard.
At the same time, ARM Macs will be much less useful for people who aren't in Apple-ecosystem. Today MacBook Pro is one of top choices for programmers and scientists - pretty much anyone who would really want a Linux, but they can't accept how awful everyday experience that brings (especially on a notebook).
And these people will probably jump to x86 devices. MS pushes Linux integration so hard - with Windows slowly becoming a dual-kernel OS - that soon it will be hard to recognize which kernel you're actually using.
So in the end Apple will probably sell significantly less devices - but with much higher margins, because they won't have to share profits with Intel and AMD.
As you mentioned yourself this depends on the software support. Hardware acceleration is meaningless if no or little software can make use of it. Apple is indeed in an excellent position there since not only do the control all the stacks, they also keep their API rather up to date to make good use of the hardware acceleration available. This is hard to compare with non-Apple systems since there the hardware, OS and software can change a lot, making similar hardware acceleration on non-closed systems like Android or PC not impossible but still very hard to achieve (especially as soon as closed software is involved that doesn't access APIs that can be extended for hardware acceleration). So non-accelerated benchmarks are still a valid base for performance comparison. As soon as hardware acceleration is used the software used need to be specified as well due to the dependency on software support.Is it even meaningful to compare performance for traditional apps anymore when so much of the transistor budget is dedicated to accelerating non-CPU and non-GPU applications?
Thanks for posting that, but I will point out that the above and one of your previous quotes represent very selective quoting. From the same earnings call:"Total operating expenses decreased by TWD2.6 billion, mainly as 5-nanometer technology moved from R&D stage to mass production during the first quarter."
- Wendell Huang -- Vice President, Finance and Chief Financial Officer/Spokesperson
Volume production at TSMC = Mass production at TSMC
I meant: when ARM Macs fully replace the x86 lineup.There's nothing stopping people from buying an Intel Mac over an iPad today if they wish. ARM Mac isn't going to change this behavior unless Apple decides to one day turn Macbooks into 2-in-1 devices, which they have said they won't do many times already.
At a more basic level, why would Apple keep a bunch of expensive inventory sitting around for months?
That's how Apple launches work. Products are available almost instantly after launch. People are camping for 2 days in front of a store to get an iPhone/a Macbook before their neighbour.At a more basic level, why would Apple keep a bunch of expensive inventory sitting around for months?
iPhone 12 isn't launching until October, and the pundits are saying that for several models, it won't even be available until November.That's how Apple launches work. Products are available almost instantly after launch. People are camping for 2 days in front of a store to get an iPhone/a Macbook before their neighbour.
Samsung does the same with their smartphones.
It's very different from what you may be used to in the PC world, when we have to wait months before some products become available.
Apple has absolutely no business in using the latest and greatest SoC they can make. No company does.In that context, having most of their A14 chips ready and sitting around in February/March, 7-9 months in advance, just doesn't make much sense. Yes I understand it's not as if they just drop in SoCs at the last minute in otherwise pre-made near-complete iPhones, but nonetheless you get the picture.
I don't understand your post.Apple has absolutely no business in using the latest and greatest SoC they can make. No company does.
A14 has a large enough performance advantage over A13 just based on the better node. Which means iPhone 12 buyers will be satisfied.
iPhone 13 will very likely use the same node, probably a SoC design similar to (based on) A14. Same can be said about the next iPad (maybe even Macs).
So it absolutely didn't make sense to "waste" all the known improvements for something that's already fine. They're saving what they can for future products.
Also, keep in mind just how many chips Apple wants to make on 5N. iPhones, iPads and potentially a big chunk of iMacs and Macbooks. So even if they wanted to delay A14 and make it slightly faster, they probably needed that node capacity for something else.
I don't understand your post.
I made no comment in that post about their performance. My point was it doesn't make sense to make these chips 8 months in advance just to sit in a warehouse.
But it's a moot argument anyway, since TSMC has already said their N5 ramp up was Q3, not Q1.
I meant: when ARM Macs fully replace the x86 lineup.
You'll have 3 groups:
1) I used a Mac and I don't care / I'm OK with ARM
2) I used a Mac but I have to stay on x86
3) I haven't used a Mac, but ARM convinced me to buy one
So yeah... I have no trouble finding arguments for (2), but I struggle to point a sensible one for (3). Hence the suspicion that market share will go down. And since Apple's margin will increase, they can totally afford it.
Sure, if Apple decides to sell Macs for less, this could bring new clients. But I look at latest iPads and I just can't find any premise of this happening.
You have to remember that Apple products are considered a luxury good in a way. So they may not want to lower the price too much.
Your arguments about efficiency and performance are true. But are they significant?
x86 laptops offer plenty of battery life today, especially current Macbooks. You're essentially moving from "more than I need between charges" to "more than I need between charges + 4 hours".
Performance (especially: performance/$) was never a decisive factor for Apple consumers.
Apple may be playing a long game - hoping that x86 tasks will migrate to cloud. But that's even more against the argument of superior chip performance (battery life as well).
High volume manufacturing is a very loose term, at least according to Intel. When they were ramping the 14nm node they had a footnote on the bottom of a slide that their definition of high volume manufacturing was 1 million die in a quarter. The die that they were using to say that 14nm was in 'high' volume production was a U/Y, meaning around 100 mm2. You can do the math on how many wafers per month is required.
TSMC continues to 'ramp' a node depending on demand/contracts. With the current capital spend rate and rumors it sounds like they are still 'ramping' the 7nm node. The 5nm ramp rate is going to be limited by EUV equipment availability. Todays news from Digitimes is that Apple is going to consume all of the 5nm capacity until the end of the year.
Apple orders to fill up TSMC 5nm process capacity
TSMC's 5nm process capacity will be fully utilized through the end of 2020, thanks particularly to strong orders for Apple's new iPad Air and upcoming iPhones, as well as the Mac, according to industry sources.www.digitimes.com
Besides building inventory for a 'mass release' Apple also uses TSMC for InFO packaging.
P.S. Anandtech has an article on Broadcom using 5nm for HPC on a double size reticle CoWoS chip. I believe that EUV cuts the maximum die size in half due to limits in lithography. This could be a problem for graphics cards from both AMD and Nvidia. I think that both are using die sizes over the EUV limit.
If someone buys a Macbook because it works well in Apple ecosystem or because it runs iPad apps (or something very similar) - he probably already own a Macbook.Each of these is legit for a different group of users, but I think the 3rd and 4th will be most important.
Again: that was not the argument.There are now many small apps that exist only in mobile form, but it can be irritating to use them on a phone -- maybe you want to cut and paste data between multiple apps, maybe you want a larger window, maybe you want a decent keyboard, maybe you're just already sitting down at your Mac -- and the ability to just use them will be, IMHO, a substantial win for the ARM Mac.
Exactly.For example I've noticed my non-techie friends who use WeChat or Viber a lot for work communication with Asia now take it for granted that they can engage in these interactions on their phone or their Mac (...)
High NA EUV lithography cuts the maximum die size in half. We aren't there yet with 5nm.
It remains unclear in High NA EUV (ASML's EXE:5000 machine) will be adopted in 3nm (increasingly unlikely) or 2nm (probable).
Yes, and I think they chose their wording carefully, as "volume production" is not necessarily the same thing as "mass production" at iPhone SoC levels. Furthermore, even disregarding that caveat, that is in Q2 already, not the Q1 you were claiming.
Maybe I missed it, but AFAIK just about nobody was claiming that TSMC was in high volume mass N5 production in Q1 2020, not even TSMC themselves.
Takes that long to be able to assemble enough iPhones for a launch.
Sounds about right. BTW, this was from August:Given that Apple sells over 200 million iPhones a year, the production process has to be capable of making 4 million a week when averaged out over the entire year. We know that Foxconn brings on extra people to increase production rates pre-launch, so it requires less than two months pre-production to satisfy all the prelaunch sales and stocking all the stores/carriers at launch.
If they can get more than they need for that initial launch/stocking great, but no matter what they will always end up with some model/color/config combos getting a few weeks out shortly after launch. That's more because they can only guess how popular the various combinations will be, and have to wait until the things start selling and see which combos are running out and which are moving more slowly than expected to make adjustments in what configs are being assembled.
Really? I had thought it was considerably less than 3 months.What I think some people are missing here is that it takes about three months for a wafer to go from the start of the process to being finished, so if you "start mass/volume production" on April 1st, the first finished wafer will come out the other end around July 1st.
When do the new iPhones come out? I don't follow this stuff.
October is the likely iPhone announcement date, but the leakers are saying that some models will not arrive until November (which is quite late).When do the new iPhones come out? I don't follow this stuff.
For fab turn around, I'm pretty sure Apple orders, at least early on, will be expedited. Fabs can typically expedite at least a good portion of your order by several weeks if you are a premier customer or are willing to pay for it.