It's really for me to believe that Intel can be competitive with TSMC in nodes again in the next 10 years.
TSMC isn't some resource-starved small company trying to compete with the Intel goliath. TSMC actually has a bigger market cap.
More importantly, TSMC is laser-focused on manufacturing, and only manufacturing, while Intel has to worry about so many different product lines in addition to node process improvements.
It wouldn't surprise me if, within the next 5 years, highend Intel chips will be manufactured by TSMC.
I don't think it'll be easy for Intel to regain competitiveness, but market cap doesn't tell you anything about how much R&D money a company can spend. Intel has double the revenue, double the net income, double the assets. Intel is profitable enough ($22 billion a year in net income) that they can spend effectively unlimited money on fab R&D if they have to.
Market cap tells enough of a story. I don't know the full financial details but the point is that Intel can't just outspend TSMC.
Intel needs to spend a lot of their R&D money on many different products, not just manufacturing, hence the point about focus.
In addition, just like this article alludes to, it's going to be rough for Intel for the foreseeable future so don't expect their revenue and net income to be as good as today.
A part of Intel‘s problem may be that this worked fine in the past. Competition ? Just use financial incentives to keep them down until you have an answer or killed it off.
Afaik this first failed with their push into the mobile space as they were not up against a single smaller competitor.
Well you can buy yourself into a market, certainly, Uber and others show how it is done. But you cant buy yourself into innovation or double the speed with double the money - SAMSUNG has been trying that with micro led and where is it?
No, this is a though one and it will take time and good people, not just $$$.You need money, but it isnt enough.
They did it once before but that tactic won't work again, as the advantage with AMD is just to big and and being fables, means neither will that hurt amd financially, nor will it limit AMD from scaling up quickly like last time. Given Intel's 10nm is not showing any significant advantage over their 14nm, things are not looking great for Intel.it looks like intel has been taken over by marketing folks unlike AMD which chose an engineer to take control of the company
The Intel machine runs on profits. Those profits are falling under attack. We’re starting to see it in drastic Xeon price drops. What’s next? Intel hopes they’ll sell lots of Tremont’s to 5G base stations. Will they? Or will Marvell Hoover those up? Even if Intel sells, will they be at historic price levels?
Even on the desktop, ARM maintains their relentless annual improvement. At some point those Chromebooks and ARM Windows devices are going to be a nice improvement over Intel. Not last year, but nicer this year, even nicer next year. And sure Intel can try to compete with 8 core Tiger Lakes — but they’ll have to offer them at competitive prices, not historic prices.
The money machine is having problems, and they look like they’re only going to get worse...
Well yes, true, but they have cash and profits in other areas. The share prices might go down and nobody likes that, those with bonuses the least, but in the end you only really care when you have to get yourself new money and they don't need that.
In other words, if intel has to, it can weather a long storm. 2024 is when they catch up? Painful, but far from a problem. Now if you said 2030, or later - that is scary for sure. But I doubt it will take them that long. I just bought a 2700X, betting a 4900x will make a fine upgrade 2 years from now - and betting it will be competitive in the market. Might even still be king of the hill. But I wouldn't bet on anything 5 years from now - intel tends to not fail for that long.
The problem for companies such as TSMC is that as they approach those smaller nodes, there’s less benefit to them, and higher costs. This is according to an AMD statement made during the 22nm process. It is truer today.
While amazingly enough. 5nm looks to be nearing fruition, 3nm is still more in the conceptual stage than anything else. There’s really nowhere to go from there at this point in time. At some point, all foundries will end up in the same place.
Who says there is less benefit at smaller nodes? We’ve heard naysayers claim this since 20nm. And it’s been nonsense for every node! You don’t think Apple and QC and SS will ship faster lower power chips at those nodes? And that people won’t buy them?
The biggest competitor of Intel is not AMD, Nvidia, QCOM or Apple, it is TSMC. I agree with you, they are in deep trouble and I am speculating that Intel is going fabless in the next decade.
TSMC and Intel share many of the same suppliers. Ultimately, whoevers on the leading edge has to foot a larger part of the enormous bill for that equipment, which gives companies slightly behind the curve (Intel, Samsung, etc) some slack to catch up.
Everyone contributed to the EUV machines being made. Each company invested billions. But Intel kind of gave up around 2010, and TSMC and ASML partnered and finally created the first EUV chip.
So no, everyone is paying the same price for that equipment. Even after its made, its still $600m per euv machine.
TSMC and ASML actually have a much better relationship than them and Intel. And since TSMC is helping design the new techniques, they actually know the best ways to use it. For the most part.
Cannon Lake (pre-10nm*, EOL, 2017) Ice Lake (10nm, launched 2019) Agilex FPGA (10nm, launched 2019) Tiger Lake (10nm+, late 2020) DG1 (10+, late 2020) Snow Ridge (10nm, late 2020) Lakefield (10nm/22nm, 2020) Ice Lake Xeon (10+, late 2020?) and Sapphire Rapids (10++, 2021?)
You missed a couple:
Water Lake (10nm++, mid 2020?) Desert Rock Stoney Rock Lake Cliff (10nm+/-. 2021?) Sewer Lake (10nm--. 2022?) Lake Water Lake (12-2nm, 2022?) Rainy Water Lake Ice Chasm (10nm-+-, 2022?) Rock McStoneyrock III (10nm++#, 2023?)
Vomit Lake, a late 2022 10nm+ backport of Comet Lake (14nm+++).
Yes, Intel HD520 will still be "powering" Intel IGPs until 2022! An IGP that can barely drive basic web browsing at 4k60 will somehow still be at Intel's front lines until 2024. Amazing work, here.
Appropriate joke on the confusing af xLake naming there. Intel needs to fire whatever monkey(s) are still pushing these - lake names. No sane person can jeep track of them anymore considering they're churning out a new lake every 6 months. At this point it'll be mentally better for all parties involved to just reset the naming and stick to simple numbers like the olden days 6700, 6750k, 6800.
Likely 10nm was necessary to squeeze in the big gen11 and gen12 integrated GPUs. The backports rumored for Rocket Lake are reduced execution units vs the 10nm.
It appears the pcie4 support required 10nm, since it was removed from the Cooper Lake design... maybe a thermal issue. I'd guess PCIE5/CXL requires 10nm for the same reason.
Intel appears to have made some nice upgrades to Agilex with 10nm.
Intel's P5900 family is 10nm, and supports extended temperature range -40 to 85c. I wonder if the power efficiency could have been met for that family without 10nm.
So, while their yields may not be great, I'm guessing Intel needs 10nm for the new features.
I suspect we'll see EUV brought in to improve 10nm yields, in addition to its use in Intel's 7nm.
No $#!& Sherlock, price/transistor and price/unit area have been rising since 22nm. That goes for everyone, not just Intel: nanoscale semiconductor physics is a bitch. More radical changes than just switching to EUV or using Cobalt metal layers are going to be needed, and GAAFETs and TFETs are only going to make the problem worse.
Plus we've already had processes abandoned outright with little or no usage. GlobalFoundries' cancellation of 7LP, Samsung quietly shuffling 7LPE under a rug, etc.
for some years, all this Qnm fight has been just a pissing contest based on vapour. when was the last time a Qnm device was actually Q nanometers in x,y,z??? or even one dimension? it's all propaganda. OTOH, shaving a nm from 10 is a larger %-age than a nm from 45, but the actual decrements, IIRC, have been at increasingly smaller (I know) %-ages and performance gains. diminishing returns, and all that. in some short time, there won't be enough atoms available to make a deterministic device.
Yeah, if 7nm to 10nm were as rigid as f/0.7 to f/1.0 then density should be 2x, so if they were serious about the measurements they'd have gone to something realistic like 8.5nm down from 10nm with what they've come up with.
22nm was their highest-yielding node of all time, so it's not a complete tragedy. Their change in position from the leader to the underdog is concerning though.
One of Intel's biggest problems will be to try to match AMD's chiplet and I/O core design - this design allows for large numbers of CPUs in a single package without having an excessively large chip. If Intel copies this design then their speed advantage of having everything on one chip vanishes, if on the other hand Intel keeps its current monolithic architecture then the yields drop due to the chip size. (Intel, if it keeps the monolithic architecture also needs to produce many more designs than AMD which uses the same chiplet design from the current Ryzen desktop chips to the EPYC Rome server chips.)
"Intel, if it keeps the monolithic architecture also needs to produce many more designs than AMD which uses the same chiplet design from the current Ryzen desktop chips to the EPYC Rome server chips."
there is precedent, albeit in software: *nix, starting at least with initial linux, couldn't decide between continuing with the monolithic kernel design against the micro-kernel (from CMU). linus opted for the 'legacy' approach, and was roasted on the innterTubes. in the end, microkernel has pretty much disappeared. we'll see, may be, whether the value of the sum of the parts exceeds the whole. Henry Ford figured out that total vertical integration, modulo excess capacity issues, yielded the maximum profit in-house. one might argue (I have more than once) that there are way more production situations which are natural monopolies. all this M&A in compute makes that pretty clear. the consumer loses.
On this point - Intel's actually talked about its investment in technology that's process-independent, like their Foveros chip stacking, so they are aware that they need to start thinking outside the wafer to stay competitive.
I don't want to excuse the massive sh*tstorm that 10nm's been for Intel, but it's getting to be a very interesting time in semiconductors as the ability to just squeeze performance out of a new process vanishes. AMD's chiplet approach - which reduces their "ask" for expensive 7nm area and helps them maximize yields (and overall strategy of throwing more cores at users top to bottom across the process stack) is one interesting solution, and I really interested in seeing what others are....
Yields there has more than one dimension too. There is the more traditional 'does everything work on the die?' approach but just as important now is 'what is the power profile of this die?'. The lower core count chips aren't lowering TDP that significantly. Part of the reason is that the lower core counts do tend to clock a bit higher but the other side of that is simply AMD is profiling for power consumption as well: higher wattage parts are being thrown into the midrange and low core count models.
Well once upon a time Intel and everyone else had their memory controller on a separate die, and it was AMD that was the first x86 vendor to change that. And now... ;-)
EMIB isn't vaporware and Intel has been shipping products using it for awhile. Their long term strategy from years ago was going to leverage it and I see no need for Intel to alter that vision as the potential gains are huge. Where things fell apart is that EMIB was to be used in conjunction with 10 nm products that never materialized due to that node's problems.
Similarly, I see Foveros being as important in the mobile space for increasing system density further. I do see Intel shipping this year as they're already sampling it seems.
"But all in all, we expect in 2020 to be fully recovered in terms of the ability to supply our customers to build inventory, we have not been able to build our own inventory, much less help customers, build their inventory, and we think both of those elements will be able to happen in 2020."
For those looking at an upside to the COVID-19 pandemic and possible economic fallout, here it is: Falling global demand might actually mean Intel might actually have the supply its customers are asking for!
To be fair, it really was Zen that started the turnaround, and both Zen and Zen+ were (are, actually since they're still selling them) made by GloFo, as is the I/O die in Zen 2.
"So for us, data centric is sort of everything but the PC that serves those markets."
Can we read between the line that Intel BnB is no more their focus? It seems like an admission that competition is now a reality for good and that Intel needs to look at other market to corner their competitors. If it is like their 5G venture, we can all be skeptical of the outcome.
"Then if you're focusing on preserving a very high market share in two core businesses - don't mistake me we're we take a very competitive view on maintaining market share and fighting for market share and our core businesses - but this data centric commentary that we're going to is real, the growth is real."
They just admitted defeat... This means, don't expect Intel to win back their node advantage, it is over.
"So if you saw the NYT this morning, there's really a nice piece on what Bob is doing to change the culture (Bob Swan, our CEO is going to change the culture as a company) to make it more focused on how do we execute and compete in a rapid, transparent one Intel way bringing all of our capabilities to tackle these new markets."
Yeah, we are seeing it... Bob is a finance guy and only care about numbers... and he is not an engineer. If Intel doesn't own the CPU business, they are not Intel anymore. Their extracurricular adventures have always been a disaster and I am not expecting less from them than a disaster.
"One of the things that happens as you enter a new market is [that] you don't have all the learnings. You don't have all the answers. You're not the expert in that space. Other people have experiences, very valuable [experiences], [that] can make you faster, and a better competitor."
Basically, they are going to buy their way in other markets? What is the plan? Buying Qualcomm? Man this is ridiculous... It is a good interview, but it is ridiculous.
"So the cost that you're absorbing, starting in particularly 2021, you’ve got this intersection of the performance of 10nm, the investment in 7nm, and we’re also well into starting the investment in 5nm” said Mr. Davis
I wonder who he thinks will be absorbing that cost? It's not like Intel aren't sitting on a pile of cash from price gouging for the last 10 or so years. Now AMD has overtaken them on process node and IPC, if AMD can deliver on stability, drivers and mobile CPUs the market will tell Intel to absorb the cost itself.
"But so, you know, I feel like we're in the 10 nanometer node. It's important that we're continuing to see yield improvements ratably over the over the time period."
"They allow you, even if you might not have the CPU positioning you would like at that particular time, you know, for instance the software on Xeon, coupled with our AI, we get like 10x the performance to the competition on AI related requirements, even with differences in the CPU."
Are we talking about Intel self-made AI benchmark AdoredTV mentioned previously? Jesus, when you know a little bit, it is easy to pinpoint all the propaganda.
"You've got to merge into traffic and nobody leaves a lane for you and it's just spectacular to see this car drive, as my friends in Israel say, drives like an Israeli! So the first 15 or 20 minutes of it, you're, you know, kind of in horror, and then the last 10 minutes, as one of the analysts said to me, 'you know I was kind of bored at the last 10 minutes', which is the highest compliment I can pay this technology. "
The Nvidia way of autonomous vehicle technology... and born for failure...
There is something in Canada called snow... just to let you know... good luck with that... (sarcasm)
Remind me again ...why does anyone still buy Intel, just stop it already and it will quickly mean they make changes and fast nothing like a minus on the books to focus the shareholders!
Desktop PC's have amazing performance here in 2020, but despite all of their advantages (large screens, high core counts, and expandability), they still suffer from one major limitation - I cannot carry them around with me. There are many great Intel powered laptops that can serve as both a desktop and mobile PC, allowing me to work from literally anywhere I need to be.
My weapon of choice for the last 13 years has been the 15" MacBook Pro. There is no other device that can replace what this laptop can do. Windows, Mac, Android, iOS, and web development - all in a sturdy anodized aluminum chassis that takes loads of abuse. I lean heavy on these machines to get things done and they have never, ever, failed me. For anyone who uses their laptops for a career, it is impossible to move away from what you KNOW works.
This is why Intel is able to charge high prices for underperforming products: unsophisticated and uninformed customers and those who find it physiologically difficult to accept they've been ripped off for years and rationalize buying Intel yet again.
>Intel intends to retake its former foundry competitive advantage when it launches its 5nm processor node, which even on the best estimates is 2024 at the earliest (and likely later),
I will have to look up what is Intel planing to do with 5nm. Intel 7nm is aiming at late 2021 or effectively 2022. So 2024 is actually within 2 years cadence.
As far as I am aware TSMC 3nm is still aiming at 2022, but it will not be GAA, which is something Samsung is *aiming* to launch. GAA is now scheduled for TSMC 2nm in 2024. ( Node number are pretty much meaningless so just read it as a generation of improvement ).
I seriously doubt Intel will regain the crown with their 5nm.
If Intel does their 7nm process (which I assume is equivalent to TSMC’s 5nm process) in 2022, then it would seem that they’d be two years behind TSMC (as it’s said that the Apple A14 will ship on TSMC 5nm in bulk in 2020), so it’s hard to see how they could close that 2-year gap and regain the crown within just one process generation.
Not quiet that way. Intel will start its 7 nm (P1276) officially in 4Q21 with Xe HPC-GPGPU "Ponte Vecchio" (and maybe Granite Rapids follows in 1HY22 also in 7 nm; Sapphire Rapids in 1HY21 will be most likely 10nm+++). Apple's SoCs are no direct competition for Intel. AMD will ship 5 nm products not before 2HY21 (Zen4-based). In 2020 they will only use TSMCs 7 nm.
Apple's SoCs are very much direct competition for Intel. Not only are the iPad Pro series eating away at the lower end of the notebook market, they're coming increasingly close to supplanting Intel in the lower-end of the MacBook range. 5nm may not be the definitive tipping point, but it's probably close.
Keep in mind Intel's "original" 7nm ( We do not know if the spec has changed ) is closer to TSMC 3nm density, somewhere between 5nm+ and 3nm. So 2024 is still doing quite good.
And as AnGe85 have said, the node is different. Although it is best to compare the best node from each side. All leading TSMC node are low power designed for Mobile Phones, higher power version tends to come a little later. With that said, Intel is also shipping low power node first to Laptop. so the comparison is still very much relevant.
They already were, too. Intel tripped over themselves trying to get a slice of the tablet market, but they literally couldn't pay their way into it - and Apple's SOCs in their iPads are a good chunk of the reason why not.
"it will create future products with back-porting and back-porting opportunities in mind"
pure garbage and empty talk. backport to what? when you cant even ship existing products, intel sillicon shortage has been apparent to us for 3 years, never mind them, while they have consistently refused to expand capacity, because its company run by the sales dept.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
92 Comments
Back to Article
senttoschool - Wednesday, March 4, 2020 - link
It's really for me to believe that Intel can be competitive with TSMC in nodes again in the next 10 years.TSMC isn't some resource-starved small company trying to compete with the Intel goliath. TSMC actually has a bigger market cap.
More importantly, TSMC is laser-focused on manufacturing, and only manufacturing, while Intel has to worry about so many different product lines in addition to node process improvements.
It wouldn't surprise me if, within the next 5 years, highend Intel chips will be manufactured by TSMC.
senttoschool - Wednesday, March 4, 2020 - link
really hard* for meGuspaz - Wednesday, March 4, 2020 - link
I don't think it'll be easy for Intel to regain competitiveness, but market cap doesn't tell you anything about how much R&D money a company can spend. Intel has double the revenue, double the net income, double the assets. Intel is profitable enough ($22 billion a year in net income) that they can spend effectively unlimited money on fab R&D if they have to.senttoschool - Wednesday, March 4, 2020 - link
@GuspazMarket cap tells enough of a story. I don't know the full financial details but the point is that Intel can't just outspend TSMC.
Intel needs to spend a lot of their R&D money on many different products, not just manufacturing, hence the point about focus.
In addition, just like this article alludes to, it's going to be rough for Intel for the foreseeable future so don't expect their revenue and net income to be as good as today.
Qasar - Wednesday, March 4, 2020 - link
Guspaz, we all know how effective throwing money at a problem, is the way to fix something,Irata - Thursday, March 5, 2020 - link
A part of Intel‘s problem may be that this worked fine in the past.Competition ? Just use financial incentives to keep them down until you have an answer or killed it off.
Afaik this first failed with their push into the mobile space as they were not up against a single smaller competitor.
jospoortvliet - Thursday, March 5, 2020 - link
Well you can buy yourself into a market, certainly, Uber and others show how it is done. But you cant buy yourself into innovation or double the speed with double the money - SAMSUNG has been trying that with micro led and where is it?No, this is a though one and it will take time and good people, not just $$$.You need money, but it isnt enough.
sharath.naik - Tuesday, April 28, 2020 - link
They did it once before but that tactic won't work again, as the advantage with AMD is just to big and and being fables, means neither will that hurt amd financially, nor will it limit AMD from scaling up quickly like last time. Given Intel's 10nm is not showing any significant advantage over their 14nm, things are not looking great for Intel.it looks like intel has been taken over by marketing folks unlike AMD which chose an engineer to take control of the companyname99 - Wednesday, March 4, 2020 - link
The Intel machine runs on profits. Those profits are falling under attack.We’re starting to see it in drastic Xeon price drops. What’s next?
Intel hopes they’ll sell lots of Tremont’s to 5G base stations. Will they? Or will Marvell Hoover those up? Even if Intel sells, will they be at historic price levels?
Even on the desktop, ARM maintains their relentless annual improvement. At some point those Chromebooks and ARM Windows devices are going to be a nice improvement over Intel. Not last year, but nicer this year, even nicer next year. And sure Intel can try to compete with 8 core Tiger Lakes — but they’ll have to offer them at competitive prices, not historic prices.
The money machine is having problems, and they look like they’re only going to get worse...
jospoortvliet - Thursday, March 5, 2020 - link
Well yes, true, but they have cash and profits in other areas. The share prices might go down and nobody likes that, those with bonuses the least, but in the end you only really care when you have to get yourself new money and they don't need that.In other words, if intel has to, it can weather a long storm. 2024 is when they catch up? Painful, but far from a problem. Now if you said 2030, or later - that is scary for sure. But I doubt it will take them that long. I just bought a 2700X, betting a 4900x will make a fine upgrade 2 years from now - and betting it will be competitive in the market. Might even still be king of the hill. But I wouldn't bet on anything 5 years from now - intel tends to not fail for that long.
melgross - Wednesday, March 4, 2020 - link
The problem for companies such as TSMC is that as they approach those smaller nodes, there’s less benefit to them, and higher costs. This is according to an AMD statement made during the 22nm process. It is truer today.While amazingly enough. 5nm looks to be nearing fruition, 3nm is still more in the conceptual stage than anything else. There’s really nowhere to go from there at this point in time. At some point, all foundries will end up in the same place.
senttoschool - Wednesday, March 4, 2020 - link
This isn't TSMC's problem. They simply offload the cost to AMD, Apple, Qualcomm, etc.And this is true for Intel as well.
name99 - Wednesday, March 4, 2020 - link
Who says there is less benefit at smaller nodes? We’ve heard naysayers claim this since 20nm. And it’s been nonsense for every node!You don’t think Apple and QC and SS will ship faster lower power chips at those nodes? And that people won’t buy them?
ksec - Wednesday, March 4, 2020 - link
Um... No.eva02langley - Wednesday, March 4, 2020 - link
The biggest competitor of Intel is not AMD, Nvidia, QCOM or Apple, it is TSMC. I agree with you, they are in deep trouble and I am speculating that Intel is going fabless in the next decade.brucethemoose - Wednesday, March 4, 2020 - link
TSMC and Intel share many of the same suppliers. Ultimately, whoevers on the leading edge has to foot a larger part of the enormous bill for that equipment, which gives companies slightly behind the curve (Intel, Samsung, etc) some slack to catch up.Fataliity - Thursday, March 5, 2020 - link
Everyone contributed to the EUV machines being made. Each company invested billions. But Intel kind of gave up around 2010, and TSMC and ASML partnered and finally created the first EUV chip.So no, everyone is paying the same price for that equipment. Even after its made, its still $600m per euv machine.
Fataliity - Thursday, March 5, 2020 - link
TSMC and ASML actually have a much better relationship than them and Intel. And since TSMC is helping design the new techniques, they actually know the best ways to use it. For the most part.HorseSaddle - Wednesday, March 4, 2020 - link
Cannon Lake (pre-10nm*, EOL, 2017)Ice Lake (10nm, launched 2019)
Agilex FPGA (10nm, launched 2019)
Tiger Lake (10nm+, late 2020)
DG1 (10+, late 2020)
Snow Ridge (10nm, late 2020)
Lakefield (10nm/22nm, 2020)
Ice Lake Xeon (10+, late 2020?)
and Sapphire Rapids (10++, 2021?)
You missed a couple:
Water Lake (10nm++, mid 2020?)
Desert Rock Stoney Rock Lake Cliff (10nm+/-. 2021?)
Sewer Lake (10nm--. 2022?)
Lake Water Lake (12-2nm, 2022?)
Rainy Water Lake Ice Chasm (10nm-+-, 2022?)
Rock McStoneyrock III (10nm++#, 2023?)
jeremyshaw - Wednesday, March 4, 2020 - link
You missed one, too!Vomit Lake, a late 2022 10nm+ backport of Comet Lake (14nm+++).
Yes, Intel HD520 will still be "powering" Intel IGPs until 2022! An IGP that can barely drive basic web browsing at 4k60 will somehow still be at Intel's front lines until 2024. Amazing work, here.
name99 - Wednesday, March 4, 2020 - link
👍👏milkywayer - Saturday, March 7, 2020 - link
Appropriate joke on the confusing af xLake naming there. Intel needs to fire whatever monkey(s) are still pushing these - lake names. No sane person can jeep track of them anymore considering they're churning out a new lake every 6 months. At this point it'll be mentally better for all parties involved to just reset the naming and stick to simple numbers like the olden days 6700, 6750k, 6800.Spunjji - Friday, March 13, 2020 - link
🤣👏JayNor - Wednesday, March 4, 2020 - link
Likely 10nm was necessary to squeeze in the big gen11 and gen12 integrated GPUs. The backports rumored for Rocket Lake are reduced execution units vs the 10nm.It appears the pcie4 support required 10nm, since it was removed from the Cooper Lake design... maybe a thermal issue. I'd guess PCIE5/CXL requires 10nm for the same reason.
Intel appears to have made some nice upgrades to Agilex with 10nm.
Intel's P5900 family is 10nm, and supports extended temperature range -40 to 85c. I wonder if the power efficiency could have been met for that family without 10nm.
So, while their yields may not be great, I'm guessing Intel needs 10nm for the new features.
I suspect we'll see EUV brought in to improve 10nm yields, in addition to its use in Intel's 7nm.
Guspaz - Wednesday, March 4, 2020 - link
AMD: *laughs in TSMC*edzieba - Wednesday, March 4, 2020 - link
No $#!& Sherlock, price/transistor and price/unit area have been rising since 22nm. That goes for everyone, not just Intel: nanoscale semiconductor physics is a bitch. More radical changes than just switching to EUV or using Cobalt metal layers are going to be needed, and GAAFETs and TFETs are only going to make the problem worse.Ian Cutress - Wednesday, March 4, 2020 - link
He's not talking per wafer here, he's talking revenue and gross over the lifetime of the node.eva02langley - Wednesday, March 4, 2020 - link
Elementary Watson when 14nm lasted... 6 years?name99 - Wednesday, March 4, 2020 - link
Oh, snap!edzieba - Thursday, March 5, 2020 - link
The latter follows the former when per-chip prices need to remain stable or minimally increase to meet market demand.edzieba - Wednesday, March 4, 2020 - link
Plus we've already had processes abandoned outright with little or no usage. GlobalFoundries' cancellation of 7LP, Samsung quietly shuffling 7LPE under a rug, etc.FunBunny2 - Wednesday, March 4, 2020 - link
for some years, all this Qnm fight has been just a pissing contest based on vapour. when was the last time a Qnm device was actually Q nanometers in x,y,z??? or even one dimension? it's all propaganda. OTOH, shaving a nm from 10 is a larger %-age than a nm from 45, but the actual decrements, IIRC, have been at increasingly smaller (I know) %-ages and performance gains. diminishing returns, and all that. in some short time, there won't be enough atoms available to make a deterministic device.s.yu - Wednesday, March 4, 2020 - link
Yeah, if 7nm to 10nm were as rigid as f/0.7 to f/1.0 then density should be 2x, so if they were serious about the measurements they'd have gone to something realistic like 8.5nm down from 10nm with what they've come up with.III-V - Wednesday, March 4, 2020 - link
22nm was their highest-yielding node of all time, so it's not a complete tragedy. Their change in position from the leader to the underdog is concerning though.Gastec - Sunday, March 8, 2020 - link
Why is it concerning?Duncan Macdonald - Wednesday, March 4, 2020 - link
One of Intel's biggest problems will be to try to match AMD's chiplet and I/O core design - this design allows for large numbers of CPUs in a single package without having an excessively large chip. If Intel copies this design then their speed advantage of having everything on one chip vanishes, if on the other hand Intel keeps its current monolithic architecture then the yields drop due to the chip size.(Intel, if it keeps the monolithic architecture also needs to produce many more designs than AMD which uses the same chiplet design from the current Ryzen desktop chips to the EPYC Rome server chips.)
FunBunny2 - Wednesday, March 4, 2020 - link
"Intel, if it keeps the monolithic architecture also needs to produce many more designs than AMD which uses the same chiplet design from the current Ryzen desktop chips to the EPYC Rome server chips."there is precedent, albeit in software: *nix, starting at least with initial linux, couldn't decide between continuing with the monolithic kernel design against the micro-kernel (from CMU). linus opted for the 'legacy' approach, and was roasted on the innterTubes. in the end, microkernel has pretty much disappeared. we'll see, may be, whether the value of the sum of the parts exceeds the whole. Henry Ford figured out that total vertical integration, modulo excess capacity issues, yielded the maximum profit in-house. one might argue (I have more than once) that there are way more production situations which are natural monopolies. all this M&A in compute makes that pretty clear. the consumer loses.
name99 - Wednesday, March 4, 2020 - link
Actually microkernel powers xnu which powers every Apple OS.So runs on like a billion devices...
And every year Apple seems to try making xnu more Mach-like rather than less.
FunBunny2 - Wednesday, March 4, 2020 - link
xnu is hybrid, not microkernel. but, yes, just about the only use of 'micro' in the real world.name99 - Thursday, March 5, 2020 - link
Well, if you ignore L4 (eg the version running in Apple's Secure Enclave).Or QNX. Or Integrity.
It's a big world of compute out there. Imagining that the x86/C/Linux+Windows part of it is the entirety misses a lot of interesting stuff.
sing_electric - Wednesday, March 4, 2020 - link
On this point - Intel's actually talked about its investment in technology that's process-independent, like their Foveros chip stacking, so they are aware that they need to start thinking outside the wafer to stay competitive.I don't want to excuse the massive sh*tstorm that 10nm's been for Intel, but it's getting to be a very interesting time in semiconductors as the ability to just squeeze performance out of a new process vanishes. AMD's chiplet approach - which reduces their "ask" for expensive 7nm area and helps them maximize yields (and overall strategy of throwing more cores at users top to bottom across the process stack) is one interesting solution, and I really interested in seeing what others are....
Kevin G - Thursday, March 5, 2020 - link
Yields there has more than one dimension too. There is the more traditional 'does everything work on the die?' approach but just as important now is 'what is the power profile of this die?'. The lower core count chips aren't lowering TDP that significantly. Part of the reason is that the lower core counts do tend to clock a bit higher but the other side of that is simply AMD is profiling for power consumption as well: higher wattage parts are being thrown into the midrange and low core count models.name99 - Wednesday, March 4, 2020 - link
Intel claim they are world leaders in this, just wait, what with EMIB and Foveros!Unfortunately, like everything else, the wonders of these technologies won’t ship till some time in the indefinite future...
jospoortvliet - Thursday, March 5, 2020 - link
Well once upon a time Intel and everyone else had their memory controller on a separate die, and it was AMD that was the first x86 vendor to change that. And now... ;-)Kevin G - Thursday, March 5, 2020 - link
EMIB isn't vaporware and Intel has been shipping products using it for awhile. Their long term strategy from years ago was going to leverage it and I see no need for Intel to alter that vision as the potential gains are huge. Where things fell apart is that EMIB was to be used in conjunction with 10 nm products that never materialized due to that node's problems.Similarly, I see Foveros being as important in the mobile space for increasing system density further. I do see Intel shipping this year as they're already sampling it seems.
eva02langley - Wednesday, March 4, 2020 - link
"But all in all, we expect in 2020 to be fully recovered in terms of the ability to supply our customers to build inventory, we have not been able to build our own inventory, much less help customers, build their inventory, and we think both of those elements will be able to happen in 2020."HAHAHAHA... what a joke... that's a funny one.
sing_electric - Wednesday, March 4, 2020 - link
For those looking at an upside to the COVID-19 pandemic and possible economic fallout, here it is: Falling global demand might actually mean Intel might actually have the supply its customers are asking for!eva02langley - Wednesday, March 4, 2020 - link
On the contrary if manufacturing process is impacted.airdrifting - Wednesday, March 4, 2020 - link
I remember AMD was doing shitty when they were using Global Foundries, I am glad they switched to TSMC.sing_electric - Wednesday, March 4, 2020 - link
To be fair, it really was Zen that started the turnaround, and both Zen and Zen+ were (are, actually since they're still selling them) made by GloFo, as is the I/O die in Zen 2.Spunjji - Friday, March 13, 2020 - link
GloFo getting a leg-up from Samsung with 14nm helped a bunch there, too.eva02langley - Wednesday, March 4, 2020 - link
"So for us, data centric is sort of everything but the PC that serves those markets."Can we read between the line that Intel BnB is no more their focus? It seems like an admission that competition is now a reality for good and that Intel needs to look at other market to corner their competitors. If it is like their 5G venture, we can all be skeptical of the outcome.
eva02langley - Wednesday, March 4, 2020 - link
"Then if you're focusing on preserving a very high market share in two core businesses - don't mistake me we're we take a very competitive view on maintaining market share and fighting for market share and our core businesses - but this data centric commentary that we're going to is real, the growth is real."They just admitted defeat... This means, don't expect Intel to win back their node advantage, it is over.
eva02langley - Wednesday, March 4, 2020 - link
"So if you saw the NYT this morning, there's really a nice piece on what Bob is doing to change the culture (Bob Swan, our CEO is going to change the culture as a company) to make it more focused on how do we execute and compete in a rapid, transparent one Intel way bringing all of our capabilities to tackle these new markets."Yeah, we are seeing it... Bob is a finance guy and only care about numbers... and he is not an engineer. If Intel doesn't own the CPU business, they are not Intel anymore. Their extracurricular adventures have always been a disaster and I am not expecting less from them than a disaster.
eva02langley - Wednesday, March 4, 2020 - link
"One of the things that happens as you enter a new market is [that] you don't have all the learnings. You don't have all the answers. You're not the expert in that space. Other people have experiences, very valuable [experiences], [that] can make you faster, and a better competitor."Basically, they are going to buy their way in other markets? What is the plan? Buying Qualcomm? Man this is ridiculous... It is a good interview, but it is ridiculous.
BushLin - Wednesday, March 4, 2020 - link
"So the cost that you're absorbing, starting in particularly 2021, you’ve got this intersection of the performance of 10nm, the investment in 7nm, and we’re also well into starting the investment in 5nm” said Mr. DavisI wonder who he thinks will be absorbing that cost? It's not like Intel aren't sitting on a pile of cash from price gouging for the last 10 or so years.
Now AMD has overtaken them on process node and IPC, if AMD can deliver on stability, drivers and mobile CPUs the market will tell Intel to absorb the cost itself.
eva02langley - Wednesday, March 4, 2020 - link
"But so, you know, I feel like we're in the 10 nanometer node. It's important that we're continuing to see yield improvements ratably over the over the time period."He just admitted that 10nm is having poor yield.
name99 - Wednesday, March 4, 2020 - link
They’ve been having improved 10nm yields since 2017. By this point yield should be somewhere around 370%!Spunjji - Friday, March 13, 2020 - link
Continuous improvements = 10% > 15% > 20% > 25% > etc. etc.eva02langley - Wednesday, March 4, 2020 - link
"They allow you, even if you might not have the CPU positioning you would like at that particular time, you know, for instance the software on Xeon, coupled with our AI, we get like 10x the performance to the competition on AI related requirements, even with differences in the CPU."Are we talking about Intel self-made AI benchmark AdoredTV mentioned previously? Jesus, when you know a little bit, it is easy to pinpoint all the propaganda.
Spunjji - Friday, March 13, 2020 - link
He was! "AI related requirements" is a giveaway. Good spot.eva02langley - Wednesday, March 4, 2020 - link
"That's exactly it. It's our experience that, you know, once you get, you know, a number of strong quarters in a row it starts to nose off."Yeah, it is called competition...
eva02langley - Wednesday, March 4, 2020 - link
"Which means that we tend to miss the lower end of the market where we sell what we call our small poor products. "ROFL... did Intel just called us peasants?!!!
eva02langley - Wednesday, March 4, 2020 - link
"You've got to merge into traffic and nobody leaves a lane for you and it's just spectacular to see this car drive, as my friends in Israel say, drives like an Israeli! So the first 15 or 20 minutes of it, you're, you know, kind of in horror, and then the last 10 minutes, as one of the analysts said to me, 'you know I was kind of bored at the last 10 minutes', which is the highest compliment I can pay this technology. "The Nvidia way of autonomous vehicle technology... and born for failure...
There is something in Canada called snow... just to let you know... good luck with that... (sarcasm)
wow&wow - Wednesday, March 4, 2020 - link
How about the 245 security vulnerabilities, fixed in chip design or still firmware and/or software patching?alufan - Wednesday, March 4, 2020 - link
Remind me again ...why does anyone still buy Intel, just stop it already and it will quickly mean they make changes and fast nothing like a minus on the books to focus the shareholders!vladx - Thursday, March 5, 2020 - link
I buy Intel because I need stable drivers and AVX 512.eva02langley - Thursday, March 5, 2020 - link
Seriously.... CPUs... drivers... AVX 512... fanboy spotted.TEAMSWITCHER - Thursday, March 5, 2020 - link
Desktop PC's have amazing performance here in 2020, but despite all of their advantages (large screens, high core counts, and expandability), they still suffer from one major limitation - I cannot carry them around with me. There are many great Intel powered laptops that can serve as both a desktop and mobile PC, allowing me to work from literally anywhere I need to be.My weapon of choice for the last 13 years has been the 15" MacBook Pro. There is no other device that can replace what this laptop can do. Windows, Mac, Android, iOS, and web development - all in a sturdy anodized aluminum chassis that takes loads of abuse. I lean heavy on these machines to get things done and they have never, ever, failed me. For anyone who uses their laptops for a career, it is impossible to move away from what you KNOW works.
prisonerX - Thursday, March 5, 2020 - link
This is why Intel is able to charge high prices for underperforming products: unsophisticated and uninformed customers and those who find it physiologically difficult to accept they've been ripped off for years and rationalize buying Intel yet again.Gastec - Sunday, March 8, 2020 - link
And what exactly can this amazing MacBook Pro do that that can't be..."replaced" by other devices?Spunjji - Friday, March 13, 2020 - link
"sTaBlE dRiVeRs"Name a single CPU-related driver problem AMD have had.
AVX 512, so essential that nothing uses it.
ksec - Wednesday, March 4, 2020 - link
>Intel intends to retake its former foundry competitive advantage when it launches its 5nm processor node, which even on the best estimates is 2024 at the earliest (and likely later),I will have to look up what is Intel planing to do with 5nm. Intel 7nm is aiming at late 2021 or effectively 2022. So 2024 is actually within 2 years cadence.
As far as I am aware TSMC 3nm is still aiming at 2022, but it will not be GAA, which is something Samsung is *aiming* to launch. GAA is now scheduled for TSMC 2nm in 2024. ( Node number are pretty much meaningless so just read it as a generation of improvement ).
I seriously doubt Intel will regain the crown with their 5nm.
Guspaz - Thursday, March 5, 2020 - link
If Intel does their 7nm process (which I assume is equivalent to TSMC’s 5nm process) in 2022, then it would seem that they’d be two years behind TSMC (as it’s said that the Apple A14 will ship on TSMC 5nm in bulk in 2020), so it’s hard to see how they could close that 2-year gap and regain the crown within just one process generation.AnGe85 - Thursday, March 5, 2020 - link
Not quiet that way. Intel will start its 7 nm (P1276) officially in 4Q21 with Xe HPC-GPGPU "Ponte Vecchio" (and maybe Granite Rapids follows in 1HY22 also in 7 nm; Sapphire Rapids in 1HY21 will be most likely 10nm+++).Apple's SoCs are no direct competition for Intel. AMD will ship 5 nm products not before 2HY21 (Zen4-based). In 2020 they will only use TSMCs 7 nm.
Spunjji - Friday, March 13, 2020 - link
Apple's SoCs are very much direct competition for Intel. Not only are the iPad Pro series eating away at the lower end of the notebook market, they're coming increasingly close to supplanting Intel in the lower-end of the MacBook range. 5nm may not be the definitive tipping point, but it's probably close.ksec - Friday, March 6, 2020 - link
Keep in mind Intel's "original" 7nm ( We do not know if the spec has changed ) is closer to TSMC 3nm density, somewhere between 5nm+ and 3nm. So 2024 is still doing quite good.And as AnGe85 have said, the node is different. Although it is best to compare the best node from each side. All leading TSMC node are low power designed for Mobile Phones, higher power version tends to come a little later. With that said, Intel is also shipping low power node first to Laptop. so the comparison is still very much relevant.
KPOM - Saturday, March 7, 2020 - link
If Apple starts producing ARM Macs, then they do become “competition” for Intel, since that’s a lot of high-end computers NOT using Intel processors.Spunjji - Friday, March 13, 2020 - link
They already were, too. Intel tripped over themselves trying to get a slice of the tablet market, but they literally couldn't pay their way into it - and Apple's SOCs in their iPads are a good chunk of the reason why not.azfacea - Wednesday, March 4, 2020 - link
"it will create future products with back-porting and back-porting opportunities in mind"pure garbage and empty talk. backport to what? when you cant even ship existing products, intel sillicon shortage has been apparent to us for 3 years, never mind them, while they have consistently refused to expand capacity, because its company run by the sales dept.
Sahrin - Thursday, March 5, 2020 - link
3x the development cost, and that makes it less profitable? No.zamroni - Friday, March 6, 2020 - link
i am glad that amd ditched global foundry as primary foundry. Intel must do this tooQasar - Saturday, March 7, 2020 - link
huh ? why ? intel has their own fabs.FunBunny2 - Saturday, March 7, 2020 - link
IIRC, wasn't there fab tech the reason they were first among equals for decades?Spunjji - Friday, March 13, 2020 - link
Yup, with minor hiccups - their initial 90nm process was inferior to AMD's 90nm SOI process.YoloPascual - Thursday, March 26, 2020 - link
And we have people in the past years claiming that intel 10nm will be better than 7nm TSMC lmao