Friday, July 24th 2020
Intel 7nm CPUs Delayed by a Year, Alder Lake in 2H-2021, Other Commentary from Intel Management
Intel's silicon fabrication woes refuse to torment the company's product roadmaps, with the company disclosing in its Q2-2020 financial results release that the company's first CPUs built on the 7 nanometer silicon fabrication node are delayed by a year due to a further 6-month delay from prior expectations. The company will focus on getting its 10 nm node up to scale in the meantime.
The company mentioned that the 10 nm "Tiger Lake" mobile processor and "Ice Lake-SP" enterprise processor remains on-track for 2020. The company's 12th Generation Core "Alder Lake-S" desktop processors won't arrive before the second half of 2021. In the meantime, Intel will launch its 11th Gen Core "Rocket Lake" processor on the 14 nm node, but with increased IPC from the new "Cypress Cove" CPU cores. Also in 2H-2021, the company will launch its "Sapphire Rapids" enterprise processors that come with next-gen connectivity and updated CPU cores.
It's interesting to note that Intel was specific about "CPU" when talking about 7 nm, meaning that Intel's foundry woes only affect its CPU product stack, and not a word was mentioned in the release about the company's discrete GPU and scalar compute processors that are being prototyped and validated. This is probably the biggest hint we'll ever get from Intel that the company's dGPUs are being designed for third-party foundries (such as Samsung or TSMC), and that the Xe dGPU product roadmap is disconnected from that of Intel's fabs.
Given the delays in Intel's 7 nm foundry node, the first Intel client-segment processors based on the node won't arrive before late-2022 or 2023, which means refinements of the current 10 nm silicon fabrication node should support Intel's client-segment product stack for the foreseeable future. The first enterprise 7 nm processors will arrive by the first half of 2023. Intel also mentioned that they expect to see "one full node improvement" from a refined 10 nanometer process, which isn't surprising, given how much experience they have improving their 14 nanometer process.
The company mentioned that the 10 nm "Tiger Lake" mobile processor and "Ice Lake-SP" enterprise processor remains on-track for 2020. The company's 12th Generation Core "Alder Lake-S" desktop processors won't arrive before the second half of 2021. In the meantime, Intel will launch its 11th Gen Core "Rocket Lake" processor on the 14 nm node, but with increased IPC from the new "Cypress Cove" CPU cores. Also in 2H-2021, the company will launch its "Sapphire Rapids" enterprise processors that come with next-gen connectivity and updated CPU cores.
Intel is accelerating its transition to 10 nm products this year with increasing volumes and strong demand for an expanding line up. This includes a growing portfolio of 10 nm-based Intel Core processors with "Tiger Lake" launching soon, and the first 10 nm-based server CPU "Ice Lake," which remains planned for the end of this year. In the second half of 2021, Intel expects to deliver a new line of client CPU's (code-named "Alder Lake"), which will include its first 10 nm-based desktop CPU, and a new 10 nm-based server CPU (code-named "Sapphire Rapids"). The company's 7 nm-based CPU product timing is shifting approximately six months relative to prior expectations. The primary driver is the yield of Intel's 7 nm process, which based on recent data, is now trending approximately twelve months behind the company's internal target.Intel's post results call also revealed a handful interesting tentative dates. For starters, "Tiger Lake" is shipping in "a matter of weeks," indicating an imminent launch ahead of the "Back to School" shopping season. Next up, the company's high-performance scalar compute processor, codenamed "Ponte Vecchio" remains slated for 2021-22, and given that it's reportedly being designed for 7 nm, we have our next big hint confirmation that these dGPUs will be built on third-party 7 nm fabs. Intel did mention that the Foveros packaging technology could be further developed over the years, and its upcoming discrete GPUs could combine dies (tiles) from multiple sources, which could include its own fabs.
Given the delays in Intel's 7 nm foundry node, the first Intel client-segment processors based on the node won't arrive before late-2022 or 2023, which means refinements of the current 10 nm silicon fabrication node should support Intel's client-segment product stack for the foreseeable future. The first enterprise 7 nm processors will arrive by the first half of 2023. Intel also mentioned that they expect to see "one full node improvement" from a refined 10 nanometer process, which isn't surprising, given how much experience they have improving their 14 nanometer process.
175 Comments on Intel 7nm CPUs Delayed by a Year, Alder Lake in 2H-2021, Other Commentary from Intel Management
We are all disappointed that they are failing to bring out proper competition to AMD Ryzens but come on.
After Sandy Bridge they did 22nm and 14nm manufacturing processes, arguably the 10nm process and 7nm is somewhere in the pipeline.
In terms of CPUs, AVX2, extending execution resources and caches, adding new tech as it comes along etc. And Ice Lake has even more substantial changes than that. Current Comet Lake (which is a rehashed Skylake from 2015) is ~25% faster than Sandy Bridge with single core load at the same clock. Ice Lake is a good step faster than that, Intel's 18% has been verified to be true enough.
This is just mainline CPUs. There is the Atom line that Intel seems to be getting back to with Tremont. There is XPoint with hopefully new gen coming out at one point.
Plus there are a bunch of other things Intel does with varying degrees of success - NAND Flash and controllers, FPGAs come to mind. Packaging technologies like EMIB or Foveros. Mobile modems and 5G is something they failed at.
Unfortunately this is what happens when you have a de facto monopoly in any industry, so this is by no means unique to Intel.
They also tried to jump further than was technically possible with their process node, as they believed that they could go down a different and supposedly better router, which turned out to be a disaster in the end. However, due to management decisions, this wasn't knocked on the head early enough, instead they tried to salvage it several times over, which lead to further delays. This is how many a company has gone bust, but luckily for Intel, there wasn't much competition, so they could continue using their current node.
This doesn't even take into account all the security holes Intel has been struggling to patch, which makes the company look like they haven't even bothered to check their products properly, as they affect so many generations of processors. There will obviously always be things like this cropping up, as someone will always find a way to bypass security, but it feels like Intel has been lazy here.
Now the competition has caught up and is about to potentially supersede Intel on many levels, which is what makes Intel look bad.
We obviously have limited insight into what they're working on, but the information that is available makes it look like it's at last another 2-3 years before Intel might be able to sort all this out and be back with something truly competitive, both in terms of process node and desktop CPU performance.
However, you are right that Intel is also working on a lot of other things, although XPoint wasn't done just by Intel, as it was co-designed with Micron and the FPGA was added though an acquisition of Altera, so that wasn't developed in-house.
Maybe the loss of focus is partially what caused some of the problems in the company, there were too many BU's fighting for resources and that's why things are where they are today. I guess you also forgot about Intel's attempt at competing with ARM in the mobile phone space, which ended in a disaster, as there's no other word for it.
Is Intel a terrible company because of all this? Of course not, but it shows that even the giants can fall from grace.
Edit: Ice Lake and Tiger Lake are obviously not without issues. 10nm is still rough, core count is too low for whatever reason, marketing is doing them no favors, do not exist on desktop etc. But in terms of CPU architecture and performance, they are not in a bad place.BUs fighting for resources is normal for all corporations :)
Intel's mobile attempt is definitely not the only thing I forgot. They are dealing in too many areas.
Now we also know why Apple chose this time to switch to ARM.
On the other hand, Zen 3 will be ultra expensive to avoid pushing Intel to drop prices.
Without a reliable process node, their CPU designs aren't going to help them get out of the hole they dug themselves into.And you know this how? Obviously AMD is going to cash in on being the top dog for the time, I mean, they need the money, but ultra expensive...
Surprise, surprise...
What's hurting Intel most right now is the culture among upper management of "we're the best, we'll figure it out, we don't need outside help". Pride comes before a fall, as they say, and Intel is still prideful and still falling. The only question remaining is how long before they get fresh blood that isn't bound by the sunk cost fallacy, and is willing to throw the whole 10nm mess (and potentially 7nm too now, it seems) on the garbage heap.
Personally I would love Intel becoming noncompetitive for a couple of years or more if needed. That way it will lose it's influence on OEMs and the tech press and give AMD the room needed to become a viable competitor to Intel. Intel is using it's money and influence the last 20 or more years, to secure an 80% or more of the market, even when having worst products. We need two companies that will have a 50-50 or at least a 60-40 percent of the market, not a company owning 80% of the market, pushing to 90% in some periods, thanks to its big pockets.
We all know what a "competitive Intel" means. Full control of the OEM and server market, full influence on the tech press and benchmark suites/sites. No thanks. Let Intel shrink a little, let AMD grow a lot, just so we can hope for a duopoly that works, not a duopoly that is a monopoly in disguise.They have a very competitive Zen 2 line of models in the market that they can keep selling. The XT models where made to keep Zen 2 prices up and the 5600XT fiasco showed us that AMD's mentality is changing. They will create a mess instead of dropping the MSRP price by 7-10%.
We now also know that they don't need to fear a responce from Intel. Only a price war that they can avoid by not pushing prices down. They can still sell cheap to OEMs, but in retail they will put higher prices to drive margins up and try to create the image of being the premium brand. If they don't try to convince consumers now, that the AMD brand is in fact the premium brand, when are they going to do it? When Intel starts throwing out more 10nm models or when they finaly fix their 7nm problems(they might never manage to fix those, but I bet at AMD they build strategies based on Intel's fast recovery in manufacturing just to be ready for that possibility).It's not only that. It's the security they feel as long as OEMs stick with them, as long as most ITs keep choosing Xeons. They keep going from record revenue to record revenue. The pile of money is keep growing in front of them, hiding the train that is coming to run over them.
I was looking at those Xeons I spoke about earlier and the lower end versions of them like Bronze are hilariously awful. They are based on the old iteration of Skylake-X, but with nothing that even made those good. They're locked, have no HT and no turbo. Base is around 1.5-2 Ghz, awful arrangement just like the CPU found in the Mac Pro. Intel is seriously still selling these.
AMD has a better for most things end user, like their only lack is native AVX-512 support, but that's not where they were headed first. They tried to take Intel down from the desktop market and they did most of it. Servers could still use their old Intel gear, depending on the size and needs of the cluster, and I think it is partly a case of that. Renoir is a far better CPU than the old Intel chip kinds of laptops still use. AMD is making a phone CPU too, they're putting their goods to where end consumer wants to see. I guess OEMs keep using massive trays of leftover Intel CPUs for their NUCs and whatnot (say 7200U), and this is a part of market they control...
Supposedly, this 7nm slip is due to a defect in the process that Intel is confident they have isolated and can fix (take that claim with a scoop of salt, but the fact that they are being specific about issues with 7nm vs the dead silence from continuing 10nm issues is positive).
But the most interesting snippet is this:This tells me that the OEMs have made it very clear to Intel that unless the latter starts hitting its projected dates again, they are going to AMD. Evidently the tipping point between familiarity with Intel but having to deal with missed releases, vs AMD predictability but having to retool for it, has been reached.
Now, whether Intel actually takes that message seriously enough is anyone's guess - see my previous comments about pride and corporate culture - but it seems that Intel is finally feeling the pressure it should have felt back in 2017 when 10nm first slipped.
You can't simply take something designed for one process node and make it on someone else's process node. This is why companies have to stick to the foundry they've designed their part to made at.
There were some cross-compatibility at one point between GloFo, Samsung and someone else, but that was on something like 32 or 2x nm and still required a fair bit of extra work to move between the two foundries.
So say Intel would take anyone of their current products and move it to TSMC, they'd have to spend something like six months to just transition their designs to work with TSMC's process node, then most likely spend another couple of months to make sure the tape-out is successful and then tune that over the better part of six months to a year before they're running at a decent production rate. So no, Intel wouldn't hit the ground running, unless they designed a new chip from more or less scratch, specifically for the TSMC process node, which might be quicker than trying to transplant a current design based on Intel's process node.
In all fairness, Intel has already made some products on various TSMC process nodes, so they might be able to do things a little bit quicker because of that, but it's still going to be more or less a year from start to finish.The question that remains unanswered though is this: Does TSMC have enough capacity to manufacture for Intel? Assuming they would go with TSMC.
Samsung might have the capacity, but I'm not aware of anything Intel has ever made in one of Samsungs foundries.
Can ya smell what kinda milk da rock be cookin ...:roll:...:D...:peace:
That was not meant as a realistic option, just an out of the blue example . Intel will likely figure out the manufacturing process sooner or later. By all indications there is enough money to bleed until they do.
I think it is due to company culture. These discrete product segmentations cost Intel to skip mixed loads altogether. For instance, there was a time when Intel could halt the AMD surge into the streamer pc market. No one knew AMD back then, it was all singular workloads until concurrently cpu encoding at the same time showed up. The AMD multithreading prowess wasn't taken for serious.
It was their esram featured chips that didn't get the greenlight from the higher ups. They were much cheaper(market buffered) and possibly a better candidate than intel's mainstream cpus to extend feature support. If intel somehow made the slight performance, they would have a line up based on esram feature level(they said time and again they had 4 times the amount projected for a measly 4 core cpu, they could have extended to its own distinction).
I suspect Intel's 7nm needs to be lined up against TSMC's "5".
It's not even remotely as bad as presented.
Well, people are jumping ship.
TSMC's 7nm are in a much better condition than Intel's 10nm. People can expect 5nm from TSMC to also be in good shape, but after that lattest announcement from Intel, 7nm is a huge questionmark. not even a "not so good node". It is a "huge question mark node".
This might devolve into gpu trolling stereotypes which steam charts are playing the joke in question. It won't go so lightly however. 860 is both cheap and dominant in a way fake virtual accounts in chinese cyber cafes cannot tip the balance.
While AMD has made great advances in the DIY space, they still need to gain more ground in mobile, OEM desktop, and enterprise. Ideally, AMD needs to get as close as possible to controlling 50% of the x86 T.A.M. in order to ensure that their current success isn't just a temporary salient that can be rolled back just as quickly.
I'm sure everyone here as enjoyed the spoils of the new competition, personally, five or six years ago, I didn't expect to have an 8, 12 and 16 core mainstream CPUs available at the prices for which they're currently available. If we want this trend to continue, and this competition to be a permanent fixture of the PC market, then I think we should be in favor of AMD having a few more years of success at Intel's expense.