Intel Reportedly Signs Deal with TSMC To Outsource 3 nm Production

[Uskompuf]

We recently reported on Intel's goal to launch their 7 nm node in 2023 which would put them on track to directly compete with TSMC's 3 nm node. It would seem like Intel has partially accepted defeat according to a recent DigiTimes report which alleges that Intel has signed a deal with TSMC to mass-produce 3 nm processors starting H2 2022. The report goes on to detail an arrangement where TSMC manufacturers the bulk of Intel processors with in-house production expected to continue albeit at lower quantities. This arrangement would also see Intel and TSMC cooperate on 2 nm products. If this deal turns out to be real Intel would become TSMC's second-largest customer after Apple.



View at TechPowerUp Main Site

 

[hat]

Not a bad deal for Intel. Not only do they get to circumvent their own shitty node, but they also get to knock AMD's production capability down a few notches. Still, TSMC can only do so much, so... can we expect availability issues from Intel as well?

 

[john_]

Not good. Don't get me wrong. AMD needs more time and to make much more money to become a true competitor to Intel and Nvidia. It's too early to see Intel gaining again an advantage over AMD.
And this time I wouldn't expect a resurrection of AMD after 15 years. What we call outrageous prices today, could become the normal in 10-15 years, because of lack of competition. The only hope of competition would be that between the ARM architecture and the X86 architecture. Nvidia vs Intel. And that's terrifying at least.

 

[ratirt]

So if Intel will make chips at TSMC's supervision and use TSMC's fabs becoming one of the largest customers, what will happen with Intel's fabs? Are they going to sell those like AMD did with GF?

 

[hat]

So if Intel will make chips at TSMC's supervision and use TSMC's fabs becoming one of the largest customers, what will happen with Intel's fabs? Are they going to sell those like AMD did with GF?

The smart move would be for them to fix their fabs and get them running properly, while still using TSMC, even if all they make is chipsets or something, just to screw everyone else out of some production capacity.

 

[s3thra]

The smart move would be for them to fix their fabs and get them running properly, while still using TSMC, even if all they make is chipsets or something, just to screw everyone else out of some production capacity.

Agreed. I'd say Intel isn't exactly thrilled about having to go to TSMC for manufacturing, though for them to get through the next few years, competitively they've had to concede for now. It's better for them to swallow their pride temporarily, and in the meantime they can re-tool their fabs and continue with their R&D for producing smaller nodes.

While manufacturing with TSMC, I'd speculate they'd want to get their fabs back to the cutting edge down the track again. Obviously what they are doing at the moment isn't working, so this might just buy them some time.

 

[watzupken]

Not good. Don't get me wrong. AMD needs more time and to make much more money to become a true competitor to Intel and Nvidia. It's too early to see Intel gaining again an advantage over AMD.
And this time I wouldn't expect a resurrection of AMD after 15 years. What we call outrageous prices today, could become the normal in 10-15 years, because of lack of competition. The only hope of competition would be that between the ARM architecture and the X86 architecture. Nvidia vs Intel. And that's terrifying at least.

My take is that Intel was always dominating in the CPU space because they have the advantage of both the fab and CPU architechture. Now the fab advantage is gone because they are on the same node as everyone else, i.e. AMD, Nvidia/ARM. So I won't say that Intel will gain any advantage from this move. 3nm is still work in progress, or in the very early stage. So they will not be able to utilize this likely until 2023. Moreover, I don't believe Intel will farm out their high end CPU fab to TSMC.

 

[john_]

My take is that Intel was always dominating in the CPU space because they have the advantage of both the fab and CPU architechture. Now the fab advantage is gone because they are on the same node as everyone else, i.e. AMD, Nvidia/ARM. So I won't say that Intel will gain any advantage from this move. 3nm is still work in progress, or in the very early stage. So they will not be able to utilize this likely until 2023. Moreover, I don't believe Intel will farm out their high end CPU fab to TSMC.

Intel maintains the advantage of high capacity production in it's fabs. AMD is fighting for wafers on the other hand. TSMC is a business. It will give capacity to the higher bidders. That's Apple and in case this rumor is true, Intel. It will not cut AMD as a customer, because others will be paying more, but still AMD will be unable to secure the capacity it needs. So the ceiling of success for AMD will remain low. No matter the quality of it's products.

Right now Intel is enjoying huge capacity and close bonds with OEMs. It is at the same or better level than AMD in architecture and only loses in manufacturing process, having to play mostly at 14nm.
Intel going at TSMC means that it can have more advantages compared to today. It can build certain CPU models at TSMC, grab the top spots in certain performance charts, while also selling what it can make in it's fabs to cover other segments of the market. The "TSMC made CPUs" will be enough to create a very possitive image of Intel's products in general. The same way a 3090 helps Nvidia to also sell 1650s for example, those "TSMC made" Intel products, will be selling 10nm and 14nm "Intel made" CPUs, even if they are inferior in every way. At the same time it grabs capacity from TSMC that could go to AMD. As we can see all those last quarters, AMD is selling everything it makes and still it is unable to cover demand. More capacity from TSMC could change this in favor of AMD. Intel grabbing that capacity will limit AMD's future success. Especially with OEMs. 2023 looks far, but it is not.

 

[Mouth of Sauron]

Not good. Don't get me wrong. AMD needs more time and to make much more money to become a true competitor to Intel and Nvidia. It's too early to see Intel gaining again an advantage over AMD.
And this time I wouldn't expect a resurrection of AMD after 15 years. What we call outrageous prices today, could become the normal in 10-15 years, because of lack of competition. The only hope of competition would be that between the ARM architecture and the X86 architecture. Nvidia vs Intel. And that's terrifying at least.

Sad, but true.

Under years of Intel dominance, what we were left was Single-Thread optimized software, low core number and ignoring progressive initiatives like HSA - now Intel has its own HSA initiative, they suddenly came to conclusion that i3 could actually have more than 2 cores and software is generally optimized better for multicore workloads.

We have better Intel thanks to AMD.

But it's still too early for factual rooting for Intel, they still have too much cash, too weak architecture, same mentality...

I seriously doubt that i3-10100 would be any different from, say, i3-7010 if there weren't for cheap Ryzen alternatives.

For TSMC-Intel relations and possible 2022 production, I would be careful with any prognosis - I don't say it won't happen, but there are some conditions... Where I work, we do relatively lots of contracts. Cruel capitalism and all, but it may be actually cruel only *after* we calculate in that new contract with another company also takes at least 6 months for validation/QA/other-papers/production-tests, so we tend to favour more reliable customers and not venture into short-term agreements. Just my 0.02g.

 

[Divide Overflow]

Intel resorts to copying AMD yet again.

 

[prtskg]

Samsung might gain some customers from TSMC too.

 

[m4dn355]

Intel resorts to copying AMD yet again.

You mean gluing contracts together

 

[RandallFlagg]

I think this report is pretty sketchy on details, and should be viewed with some skepticism. I think a lot of the AMD vs Intel talk is really overblown as well, as it is the perspective of people who frequent these forums and primarily think about desktop PCs.

Intel's main competitor is no longer AMD, it's the ARM architectures primarily at Amazon but also coming in from Apple. AMDs growth has always been capped by the capacity and wafer allocation of TSMC, and TSMC is not even close to dedicated to AMD. Apple, Qualcomm - both of which dwarf AMD - along with an army of dozens of other customers also have to be serviced by TSMC. Add in that half of TSMCs fabs are 12nm+, and there's never been any realistic chance of AMD taking really huge market share from Intel in the short to intermediate term. Maybe in 5-6 years, when TSMC has a few more 5nm and below fabs online, but not right now.

This all showed up in AMDs earnings. Their main growth was +176% in embedded, which is another way of saying gaming consoles. In server/client, they were +18%. The overall market was +26%.

Intel's shipments in client went up 33% in Q4.

The obvious consequence of those numbers is that Intel gained market share - if simply no other reason than that they were the only ones capable of ramping production sufficiently to meet demand.

The big hit to Intel in Q4 was not even server chips overall, which also went up 9% for the year. It was very specific, in the cloud compute area, and the big hit seems to be from Amazon Graviton 2 which is a 7nm 30B transistor 64 core chips made specifically for virtualization. The other aspect is that Intel had a blowout quarters in Q2 and Q3, driven by servers, so without new processors to sell they are once again competing with themselves.

With Intel for 2021 we have 10nm (equiv to TSMC 7nm) expanding into more laptops and servers, and we have Intel 7nm (equiv to TSMC 5nm) coming online at end of year for AL. AMD will not have 5nm until Zen 4 maybe in 2022, and last report I read was that Apple had 100% of TSMC 5nm capcity booked for 2021.

What this all boils down to is that yes, production capacity matters, and AMD doesn't have it.

 

[ODOGG26]

I will believe it when its official. As of right now it's just smoke..

 

[cst1992]

Does this mean their 7nm is dead already? Or are they thinking about leapfrogging to 3nm after 7nm is released in 2023?
One thing is for sure: if AMD manages to maintain its lead with 7nm and then 5nm by the time Intel's 7nm is out, they'll have the faster CPUs till 2023.

 

[RandallFlagg]

Does this mean their 7nm is dead already? Or are they thinking about leapfrogging to 3nm after 7nm is released in 2023?
One thing is for sure: if AMD manages to maintain its lead with 7nm and then 5nm by the time Intel's 7nm is out, they'll have the faster CPUs till 2023.

AMD doesn't make chips, TSMC does that for them.

TSMC 5nm is roughly equal to (actually, less dense than) Intel 7nm.

Intel is expected to have 7nm Alder Lake late 2021.

Apple has 100% of TSMC 5nm production occupied for 2021.

This means AMD will not have an equivalent process node chip (TSMC 5nm) to Alder Lake until 2022 at the earliest.

 

[Arc1t3ct]

This rumor is complete BS.

 

[cst1992]

@Arc1t3ct How do you know?

 

[TumbleGeorge]

Intel is expected to have 7nm Alder Lake late 2021

LoL! 7nm not before Lunar Lake in 2023. What you smoke

 

[RandallFlagg]

LoL! 7nm not before Lunar Lake in 2023. What you smoke

Intel has resolved its 7nm chip woes, 12th gen Alder Lake CPU lineup on track for 2H 2021

Intel gave analysts an update on the company's 7nm technology and roadmap, citing "strong progress" regarding this manufacturing node.

www.dsogaming.com

 

[TumbleGeorge]

Intel has resolved its 7nm chip woes, 12th gen Alder Lake CPU lineup on track for 2H 2021

Intel gave analysts an update on the company's 7nm technology and roadmap, citing "strong progress" regarding this manufacturing node.

www.dsogaming.com

LoL. This website and title of article make me feel strange. "Its chip" and "its 7nm" are different things between them has ~2 years. Will see who is more drunken, 2H 2021 is so close.

 

[Caring1]

Intel has resolved its 7nm chip woes, 12th gen Alder Lake CPU lineup on track for 2H 2021

Intel gave analysts an update on the company's 7nm technology and roadmap, citing "strong progress" regarding this manufacturing node.

www.dsogaming.com

Didn't they say the same thing about 10nm?
We're making strong progress. lol.
It's all smoke to placate investors.

 

[JayN]

Intel is expected to have 7nm Alder Lake late 2021.

Alder Lake is 10esf in 2021.
Meteor Lake was supposed to be Intel 7nm in 2022.
At one time there were leaks that the P5900 successor family, Grand Ridge, would have Intel 7nm Gracemont cores in 2021. I'd guess that would be a good candidate for TSM processing, since Intel has built Atom chips there previously.

 

[TumbleGeorge]

Meteor Lake was supposed to be Intel 7nm in 2022.

If Intel really changed his minds and are ready with work progress... But in known(rumored?) plans Meteor Lake will be also on 10nm like Alder Lake. First on 7nm will be Lunar Lake in 2023.