More pie in the sky for Intel. One day they might surprise us all and actually release competitive, innovative products, eh? Even lowly Apple booted them! That had to hurt.
You do realize they will move all their systems to their in-house Mx based designs. When Apple announced it, they told this transition will be over a period of 2 years.
But then Intel bought up a bunch of 3nm and Apple changed next years 3nm macs to 4nm...LOL. There is more than one way to skin a cat...LOL. IF you can get wafers, and I can easily turn them into gpus (at any node), I will have no trouble selling out your wafers, hurting both AMD/NV bottom lines, and making sure I have time to catch up before you GET the good wafers. :) Right?
It isn't like apple can go to samsung for better 3nm wafers. NV tried samsung 8 and got burned to some extent. So they're back to tsmc for a large order (still at samsung too, just not nearly as much). Let the wafer fights begin...ROFLMAO. So delay that to 3yrs for macs I guess...Cook made a major mistake here, as that might be all Intel needed if they bought enough 3nm out of the gate on 2nd batch here (and much more I'd hope to stop all 3nm launches from competition).
It’ll be weird to see Intel committing in such a major way with their IDM 2.0 strategy and have TSMC manufacture a leading edge CPU, which is what the rumour mentioned.
Intel will just buy wafers at TSMC and book capacity just so that it can defeat Apple? Spend money in 2 places? You have a nice strategy. Spend so much money thinking about Apple when it is not even currently it’s biggest client. Shill harder.
This has the air of one of those "predictions" that might "come true" for entirely incidental reasons. Apple don't pre-announce what nodes they're using, so anybody claiming to know anything firm about both the intended node for a product and a change of plan is probably bullshitting.
Only a fool would see it as a good strategy for a company that spends billions on its own process tech to buy wafers from a competing foundry, for the sole purpose of bullying both its own potential customers and its competition.
Even if you’re sarcastic, Apple is replacing x86 with their in house Arm design. Though, x86 is not going anywhere it’s here to stay in other devices. Apple is not the end of x86.
The M1 is the equivalent of the A6. Remember that? The proof that "we can do this -- but we're doing it at a fairly unambitious level, just to make sure everything is line up properly".
Apple hasn't even released the equivalent of the A7 for Mac yet, let alone the relentless 20..30% annual improvement we saw after the A7...
I'm curious whether they can continue to get those kinds of gains. There's lots of room for improvement when you're catching up... now that they're at or near the top (for single core IPC at least) it might be much more difficult to get more IPC. Lots of room for clock speed improvements, etc though.
> it might be much more difficult to get more IPC. Lots of room for clock speed improvements
To some degree, clock speed sits in opposition to IPC. The lower your clockspeed, the longer your critical paths can be, which enables doing more work per cycle.
Qualcomm probably got incentive to go with Intel. And a good portion of the design might actually be portable to, say, Samsung 3GAP or even TSMC N2 if dual sourcing possible is kept in mind from the beginning, and it has to be.
Dual-sourcing could be one of the big benefits of those AI layout tools we've been reading about. If it's cheaper & easier than ever to do layout, then maybe everyone will start porting their chips to multiple processes.
No, but they're certainly the best and most prolific at it. Being as they have no real government regulations on the matter and they're free to rip off whatever trademark and copyright they see fit.
> Being as they have no real government regulations on the matter
I think there have been many credible allegations that the government actually perpetrated IP theft of strategic importance and funneled the IP to domestic producers.
There have also been many cases of individuals working abroad and taking IP from their employers or universities. I think I observed this firsthand, but could find only circumstantial evidence that wouldn't support raising such a serious allegation.
That's not how it works. They don't send Intel the source code and say "build this". What they send to Intel is more like the commands your PC would send to a 3D printer. So much reverse-engineering would be needed to go from that to a comprehensible chip design that it wouldn't be worthwhile.
Plus, anyone can shave down a die and look at it under an electron microscope, if they really want to see certain low-level details.
Makes sense. Intel is promising a very power-efficient node. This gives Qualcomm a chance to make some very efficient SoC's.
Besides, they're both yank companies, they can try begging for public funds with the justification that this kind of cooperation benefits their country, instead of those dastardly *insert nationality*
Yeah, that's a good point. A lot of fuss is made about who builds the communication infrastructure. If you *really* care about that, you might even like to have the chips fabbed domestically. The main point there is probably to ensure a steady supply.
FWIW, I made a point of buying a Netgear wifi router. Among other things, it uses a Qualcomm SoC.
Even if all Intel had to do to retake the lead is meet their own stated schedule, it would still be very challenging. Possible? Yes, definitely. But challenging. I'd give the a 50-50 chance of meeting their own schedule.
But taking the lead means more than just meeting their own schedule -- they need TSMC to also follow Intel's perception of TSMC's schedule. What if TSMC were to parlay their (1) greater economics of scale and (2) greater experience with EUV, and actually beat their current public schedule?
And of course there's the squishy in between, where maybe all the schedules play out as Intel predicts, BUT we end up in a situation where TSMC's yields are great, Intel's suck, and so while Intel technically "retakes the lead" it's with a process that they dare not deploy at volume because they'd lose money.
I'll say this, though... Intel is giving every appearance of doing the right things. Hiring back seasoned engineers and making major commitments to fight hard are all good and necessary. And even if they don't unambiguously retake the lead by 2025, if they are at least competitive, and in a position to maybe retake the lead by 2030 while still being pretty profitable, well -- that's not half bad!
Good points. When TSMC releases a roadmap, they are basically initiating talks with customers, which turn into business commitments they have to deliver. So, their confidence level needs to be pretty high, before they announce it. That leaves a reasonable chance they can even surpass it.
Intel is playing a different game than they're used to. In the past, their fab roadmaps have been largely aspirational, I think.
> so while Intel technically "retakes the lead" it's with a process that > they dare not deploy at volume because they'd lose money.
What we saw with 10 nm is they had problems with power, frequency scaling, and yields. The more technical innovations you try to cram into a new process, the more things you have to optimize and the more risk you take on. And, as we've seen, that can take a long time to work through.
> even if they don't unambiguously retake the lead by 2025, if they are at least competitive, > and in a position to maybe retake the lead by 2030
Wrong article. This is about their foundry business. If they fail to deliver on their roadmap, their own CPU business can survive that, as we've seen. But their foundry business possibly can't.
I think the foundry business can survive being merely competitive, rather than retaking the lead, because they can make price adjustments.
But then that brings us to the part of the equation that involves executive bonuses, Wall Street, and internal politics. Will lower margins from a foundry business be tolerated? Or will intel go the route of ibm, and try to grow the bottom line by shrinking costs faster than they shrink revenue?
TSMC cant output as many wafers as Intel's manufacturing network.
You know who had the first ASML EUV litho machine and did years of research and engineering that fed back to ASML so everyone else in the industry actually could use said tools? It sure wasn't TSMC.
I'm looking at this as their Plan B. It doesn't seem to involve any specific commitments to /not/ use another competitor, and if they don't have any similar agreements in place with TSMC the most likely reason would be that TSMC don't feel the need to sell access to nodes that are this far out on their roadmap.
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
59 Comments
Back to Article
ricardodawkins - Monday, July 26, 2021 - link
"as a customer of both TSMC and Qualcomm"Samsung ?
Ryan Smith - Tuesday, July 27, 2021 - link
Samsung it is. Thanks!WaltC - Monday, July 26, 2021 - link
More pie in the sky for Intel. One day they might surprise us all and actually release competitive, innovative products, eh? Even lowly Apple booted them! That had to hurt.lilo777 - Tuesday, July 27, 2021 - link
Apple "booted" Intel only from their cheapest/least powerful computers.Teckk - Tuesday, July 27, 2021 - link
You do realize they will move all their systems to their in-house Mx based designs. When Apple announced it, they told this transition will be over a period of 2 years.TheJian - Tuesday, July 27, 2021 - link
But then Intel bought up a bunch of 3nm and Apple changed next years 3nm macs to 4nm...LOL. There is more than one way to skin a cat...LOL. IF you can get wafers, and I can easily turn them into gpus (at any node), I will have no trouble selling out your wafers, hurting both AMD/NV bottom lines, and making sure I have time to catch up before you GET the good wafers. :) Right?It isn't like apple can go to samsung for better 3nm wafers. NV tried samsung 8 and got burned to some extent. So they're back to tsmc for a large order (still at samsung too, just not nearly as much). Let the wafer fights begin...ROFLMAO. So delay that to 3yrs for macs I guess...Cook made a major mistake here, as that might be all Intel needed if they bought enough 3nm out of the gate on 2nd batch here (and much more I'd hope to stop all 3nm launches from competition).
Teckk - Tuesday, July 27, 2021 - link
What water is Intel buying? Intel is manufacturing on their own fabs not at TSMC. TSMC continues as-is.Teckk - Tuesday, July 27, 2021 - link
*waferFulljack - Wednesday, July 28, 2021 - link
there's a rumour by Reuters regarding Intel DG2 will be outsourced to TSMC 'enhanced' 7nm. It's still a year away so take it with a grain of salt.Teckk - Wednesday, July 28, 2021 - link
It’ll be weird to see Intel committing in such a major way with their IDM 2.0 strategy and have TSMC manufacture a leading edge CPU, which is what the rumour mentioned.Teckk - Tuesday, July 27, 2021 - link
Intel will just buy wafers at TSMC and book capacity just so that it can defeat Apple? Spend money in 2 places? You have a nice strategy. Spend so much money thinking about Apple when it is not even currently it’s biggest client.Shill harder.
mode_13h - Tuesday, July 27, 2021 - link
You seem so heavily invested in these 3 nm rumors, what if they turn out to be wrong?Qasar - Tuesday, July 27, 2021 - link
it wont be wrong.. the all mighty jian said so !! after all, it knows everything, cant you tell by its posts ??Spunjji - Wednesday, July 28, 2021 - link
This has the air of one of those "predictions" that might "come true" for entirely incidental reasons. Apple don't pre-announce what nodes they're using, so anybody claiming to know anything firm about both the intended node for a product and a change of plan is probably bullshitting.Spunjji - Wednesday, July 28, 2021 - link
Only a fool would see it as a good strategy for a company that spends billions on its own process tech to buy wafers from a competing foundry, for the sole purpose of bullying both its own potential customers and its competition.Teckk - Wednesday, July 28, 2021 - link
Still waiting for a logical reply to comprehend in between those LOLs and other acronyms.TheinsanegamerN - Tuesday, July 27, 2021 - link
BuT tHe M1 Is So PoWeRfUl ThAt X86 wIlL bE dEsTrOyEd In AlL sOfTwArE, x86 Is DoOmEd!!!1!1!111!!!!Teckk - Tuesday, July 27, 2021 - link
Even if you’re sarcastic, Apple is replacing x86 with their in house Arm design.Though, x86 is not going anywhere it’s here to stay in other devices. Apple is not the end of x86.
name99 - Tuesday, July 27, 2021 - link
The M1 is the equivalent of the A6. Remember that? The proof that "we can do this -- but we're doing it at a fairly unambitious level, just to make sure everything is line up properly".Apple hasn't even released the equivalent of the A7 for Mac yet, let alone the relentless 20..30% annual improvement we saw after the A7...
andrewaggb - Wednesday, July 28, 2021 - link
I'm curious whether they can continue to get those kinds of gains. There's lots of room for improvement when you're catching up... now that they're at or near the top (for single core IPC at least) it might be much more difficult to get more IPC. Lots of room for clock speed improvements, etc though.mode_13h - Thursday, July 29, 2021 - link
> it might be much more difficult to get more IPC. Lots of room for clock speed improvementsTo some degree, clock speed sits in opposition to IPC. The lower your clockspeed, the longer your critical paths can be, which enables doing more work per cycle.
Spunjji - Wednesday, July 28, 2021 - link
Yes, some people say silly things. You should pay less attention to them, and definitely not set your opinions in simple opposition to them.Spunjji - Wednesday, July 28, 2021 - link
Must be weird moving through the world only ever looking backwardsTeckk - Wednesday, July 28, 2021 - link
I’m not sure who this reply is for, the comment section is hard to navigate.SarahKerrigan - Monday, July 26, 2021 - link
Wasn't expecting this, but it makes a certain amount of sense - if Intel can execute.shabby - Monday, July 26, 2021 - link
Did someone forget to tell qualcomm how long it took intel to go from 14nm down to 10? They'll be waiting a looooooooooong time for that 20a process.dotjaz - Tuesday, July 27, 2021 - link
Qualcomm probably got incentive to go with Intel. And a good portion of the design might actually be portable to, say, Samsung 3GAP or even TSMC N2 if dual sourcing possible is kept in mind from the beginning, and it has to be.mode_13h - Tuesday, July 27, 2021 - link
Dual-sourcing could be one of the big benefits of those AI layout tools we've been reading about. If it's cheaper & easier than ever to do layout, then maybe everyone will start porting their chips to multiple processes.Sahrin - Monday, July 26, 2021 - link
Whelp, that's the end of Qualcomm.gdansk - Monday, July 26, 2021 - link
I'm sure Qualcomm is hedging their bets. They've been playing both Samsung/TSMC for years now and recently bought another design team.Sahrin - Monday, July 26, 2021 - link
Giving Intel your IP is like having a child molester babysit your kids and leaving them a bottle of Viagra.TheinsanegamerN - Tuesday, July 27, 2021 - link
Since when is Intel Chinese?dotjaz - Tuesday, July 27, 2021 - link
Since you become a racist, so I guess forever then.Spunjji - Wednesday, July 28, 2021 - link
Is the implied hypothesis that only Chinese companies steal IP? You sweet summer child.JKflipflop98 - Friday, July 30, 2021 - link
No, but they're certainly the best and most prolific at it. Being as they have no real government regulations on the matter and they're free to rip off whatever trademark and copyright they see fit.mode_13h - Saturday, July 31, 2021 - link
> Being as they have no real government regulations on the matterI think there have been many credible allegations that the government actually perpetrated IP theft of strategic importance and funneled the IP to domestic producers.
There have also been many cases of individuals working abroad and taking IP from their employers or universities. I think I observed this firsthand, but could find only circumstantial evidence that wouldn't support raising such a serious allegation.
mode_13h - Tuesday, July 27, 2021 - link
That's not how it works. They don't send Intel the source code and say "build this". What they send to Intel is more like the commands your PC would send to a 3D printer. So much reverse-engineering would be needed to go from that to a comprehensible chip design that it wouldn't be worthwhile.Plus, anyone can shave down a die and look at it under an electron microscope, if they really want to see certain low-level details.
Sahrin - Tuesday, July 27, 2021 - link
This has to be the most deluded post I've ever seen.Unashamed_unoriginal_username_x86 - Tuesday, July 27, 2021 - link
Charcoal calling porcelain black. Why would anyone fab at Samsung if they get your IP when you fab with themmode_13h - Wednesday, July 28, 2021 - link
Please explain.Spunjji - Wednesday, July 28, 2021 - link
It's accurate, so noSpunjji - Wednesday, July 28, 2021 - link
That's not really how any of this works.arashi - Monday, July 26, 2021 - link
Qualcomm Ericsson RIP.Wereweeb - Tuesday, July 27, 2021 - link
Makes sense. Intel is promising a very power-efficient node. This gives Qualcomm a chance to make some very efficient SoC's.Besides, they're both yank companies, they can try begging for public funds with the justification that this kind of cooperation benefits their country, instead of those dastardly *insert nationality*
mode_13h - Tuesday, July 27, 2021 - link
Yeah, that's a good point. A lot of fuss is made about who builds the communication infrastructure. If you *really* care about that, you might even like to have the chips fabbed domestically. The main point there is probably to ensure a steady supply.FWIW, I made a point of buying a Netgear wifi router. Among other things, it uses a Qualcomm SoC.
flgt - Tuesday, July 27, 2021 - link
This feels like a paid-for press release. A few little R&D projects and no real commitments. Easy money for Qualcomm.mode_13h - Tuesday, July 27, 2021 - link
They're reporting on a press release, so yeah.Blastdoor - Tuesday, July 27, 2021 - link
Even if all Intel had to do to retake the lead is meet their own stated schedule, it would still be very challenging. Possible? Yes, definitely. But challenging. I'd give the a 50-50 chance of meeting their own schedule.But taking the lead means more than just meeting their own schedule -- they need TSMC to also follow Intel's perception of TSMC's schedule. What if TSMC were to parlay their (1) greater economics of scale and (2) greater experience with EUV, and actually beat their current public schedule?
And of course there's the squishy in between, where maybe all the schedules play out as Intel predicts, BUT we end up in a situation where TSMC's yields are great, Intel's suck, and so while Intel technically "retakes the lead" it's with a process that they dare not deploy at volume because they'd lose money.
I'll say this, though... Intel is giving every appearance of doing the right things. Hiring back seasoned engineers and making major commitments to fight hard are all good and necessary. And even if they don't unambiguously retake the lead by 2025, if they are at least competitive, and in a position to maybe retake the lead by 2030 while still being pretty profitable, well -- that's not half bad!
mode_13h - Tuesday, July 27, 2021 - link
Good points. When TSMC releases a roadmap, they are basically initiating talks with customers, which turn into business commitments they have to deliver. So, their confidence level needs to be pretty high, before they announce it. That leaves a reasonable chance they can even surpass it.Intel is playing a different game than they're used to. In the past, their fab roadmaps have been largely aspirational, I think.
> so while Intel technically "retakes the lead" it's with a process that
> they dare not deploy at volume because they'd lose money.
What we saw with 10 nm is they had problems with power, frequency scaling, and yields. The more technical innovations you try to cram into a new process, the more things you have to optimize and the more risk you take on. And, as we've seen, that can take a long time to work through.
> even if they don't unambiguously retake the lead by 2025, if they are at least competitive,
> and in a position to maybe retake the lead by 2030
Wrong article. This is about their foundry business. If they fail to deliver on their roadmap, their own CPU business can survive that, as we've seen. But their foundry business possibly can't.
Blastdoor - Wednesday, July 28, 2021 - link
I think the foundry business can survive being merely competitive, rather than retaking the lead, because they can make price adjustments.But then that brings us to the part of the equation that involves executive bonuses, Wall Street, and internal politics. Will lower margins from a foundry business be tolerated? Or will intel go the route of ibm, and try to grow the bottom line by shrinking costs faster than they shrink revenue?
mode_13h - Thursday, July 29, 2021 - link
> I think the foundry business can survive being merely competitive,> rather than retaking the lead,
Sure, but broken promises can quickly get expensive.
Spunjji - Wednesday, July 28, 2021 - link
Being a close second to a TSMC that's executing the way it currently is would also not be a bad result by any means.JKflipflop98 - Friday, July 30, 2021 - link
TSMC cant output as many wafers as Intel's manufacturing network.You know who had the first ASML EUV litho machine and did years of research and engineering that fed back to ASML so everyone else in the industry actually could use said tools? It sure wasn't TSMC.
Blastdoor - Thursday, August 5, 2021 - link
Sure, and the Greeks invented Democracy. There's a limit to how long you can rest on your laurels.mode_13h - Tuesday, July 27, 2021 - link
At the first hint of schedule-slip by Intel, I'll bet Qualcomm is going to start hedging their bets by pursuing a Plan B with TSMC or Samsung.Spunjji - Wednesday, July 28, 2021 - link
I'm looking at this as their Plan B. It doesn't seem to involve any specific commitments to /not/ use another competitor, and if they don't have any similar agreements in place with TSMC the most likely reason would be that TSMC don't feel the need to sell access to nodes that are this far out on their roadmap.Freeb!rd - Tuesday, July 27, 2021 - link
Intel probably offered "buy one get 5 wafers free" to get Qualcomm in the door...CyborgAlienRay - Tuesday, July 27, 2021 - link
Not that it's going to matter but, hold onto your wallet... Prices will sky rocket!!!regsEx - Sunday, August 1, 2021 - link
Do we know SMIC "14 nm" and "7 nm" density?