Speculation: Intel will become fabless

Page 7 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

With the loss of its manufacturing lead, will Intel become fabless?

  • Yes, Intel is a product designer at heart, and they will seek a more flexible fabless model.

    Votes: 24 13.4%
  • No, manufacturing is integral to Intel, and they will continue to invest to stay competitive.

    Votes: 155 86.6%

  • Total voters
    179

soresu

Platinum Member
Dec 19, 2014
2,617
1,812
136
Don't even make a comparison between China and India since the latter is not even close to technological competence until another 2 decades while the former is already superior to America in some cases such as telecommunications equipment ...

China's independently developed x86 cores such as the Zhoaxin KX-7000 has IPC at least comparable to the Haswell architecture ...
That linked article showed a single Geekbench score inferior to Cortex A72 scores I have seen - even if it was a particularly low clock the GB score was running at, a high IPC at low clock does not guarantee that it will clock high enough to be competitive with Haswell, let alone Zen2.

Besides which, if it is so good then why did they even bother with the Hygon/Dhyana joint venture in the first place?

As to India, they are developing R5 cores in the Shakti program at the Madras University (IIT Madras) - I don't know where you get this strange 2 decade mumbo jumbo from, India produces a fair amount of intelligent individuals that enter the tech space, I'm fairly sure that Raja Koduri numbers among them, as are/were likely many others at AMD and Intel over the years.

Don't make the mistake of seeing the worst parts of the country in the media and judging the competence of the whole on that - there are some parts of China that are at least as backward I don't doubt.
 

moinmoin

Diamond Member
Jun 1, 2017
4,934
7,619
136
Could somebody with technical expertise explain in laymans terms how a company with such a dominant postition as Intel got themselves into such a predicament?
Intel built their dominance on their fab techs being ahead of all other fabs. For them it was always fabs first and silicon designs second.

From 1972 to 2011 (10 µm to 22nm) Intel had a new node at least every 2 years, so even if the silicon design improvements weren't that impressive, the steady node advancements ensured products that were essentially unrivaled in the market.

The next node, 14nm, then had both a delay and a slow roll out. Instead learning from that Intel tried to do even more at once with 10nm, with the known result. This turns out to be fatal since so far their silicon designs were inherently linked to the nodes, and now that the new node keeps being the delayed they had to make do with the older design on the older node (thus all the Skylake based variants). Node and silicon designs being kept together also meant Intel tried to make new nodes as feature-full as possible, which caused the delay with 14nm as well as the spectacular failure with 10nm. Going by their traditional biannual schedule Intel's 7nm node should have launched back in 2017. Now the tentative launch is end of 2021, with no indication yet if it will be a slow roll out like 14nm or even a broken one like 10nm.

TSMC on the other hand, being a pure play fab, is financed and pushed mainly by the mobile market and its demand for improved models every single year. The shorter time frame means TSMC creates a lot of smaller "in between" nodes, being more gradual with introducing new tech. This allows them to juggle tech and ensure that their roadmap is always filled with new nodes offering customers improvements launching as announced.

What makes this the perfect storm: Until AMD announced their move to TSMC all the mobile market development did not have any crossover with the x86 market so there was no real way to directly compare the different nodes of the different fabs. Now Intel is lagging with nodes while AMD is embracing the yearly node cadence TSMC is offering them.
 

DrMrLordX

Lifer
Apr 27, 2000
21,583
10,785
136
China too has potential to create competitive x86 cores to capitalize these markets as well ...

No they don't. Besides, the lion's share of chip R&D in China is already being thrown at other ISAs.

It sure sounds like Intel has no intention of going fabless any time soon.

Just because Intel is throwing money at something doesn't mean that it'll work. How much do you suppose they've spent on 10nm already? Intel knows that 7nm is essentially their last stop. If it can't be made to work in an acceptable time frame, their foundry days are mostly finished. They will be permanent also-rans or foundry vassals of Samsung and/or TSMC.

I am sure the tentacles of mainland China reach into TSMC.

Nah. If there is any one country that wants to keep the PRC out of their business, it is Taiwan. TSMC is critical infrastructure for them. That being said, China has the brute force option.
 

UsandThem

Elite Member
May 4, 2000
16,068
7,380
146
Just because Intel is throwing money at something doesn't mean that it'll work. How much do you suppose they've spent on 10nm already? Intel knows that 7nm is essentially their last stop. If it can't be made to work in an acceptable time frame, their foundry days are mostly finished. They will be permanent also-rans or foundry vassals of Samsung and/or TSMC.
While they have struggled the last several years with 10nm development, and leadership changes, Intel seems to be confident that they are back on track.

Now that AMD has really given them very strong competition with Ryzen2, their fab / design plans need to be on target (and on time). Owning their own fabs are part of what makes them so much more profitable than AMD, so it would really have to be going bad for them to go fabless.

From the article I linked to in my last post, they seem to think they are back on track. Of course, nobody outside of Intel will know if this product confidence is legit (or smoke and mirrors) until product targets on their product road map arrive.
While it might seem doubtful that 10nm would be a one-time hiccup and the company can regain its ability to deliver on a regular cadence at 7nm, there are some arguments in favor of it. Intel says that it has learned several lessons from 10nm. It has reduced the density scaling target and the introduction of EUV should be beneficial to yield learning, as Intel has said that multiple patterning has been one of the big challenges for 10nm. It has also allowed Intel to vastly simplify the 7nm design rules.
Bob Swan said that it expects to maintain that schedule “at least for the next few nodes”, and already expressed his confidence in 5nm (which should hence launch in early 2024): "We are also well down the engineering path on 5-nanometer."

Or as he put it: "So, we're investing to recapture process leadership going forward".
 

DrMrLordX

Lifer
Apr 27, 2000
21,583
10,785
136
Intel has been "on track" since about 2017, I think. And frankly, I do not think they will catch TSMC at this point. Not until process development bottoms out.
 
  • Like
Reactions: Lodix

UsandThem

Elite Member
May 4, 2000
16,068
7,380
146
Intel has been "on track" since about 2017, I think. And frankly, I do not think they will catch TSMC at this point. Not until process development bottoms out.
From the articles I read in the past, I don't think they've been on target with their roadmaps, and had to adjust them:
https://www.tomshardware.com/news/intel-cpu-10nm-earnings-amd,36967.html
https://www.anandtech.com/show/12693/intel-delays-mass-production-of-10-nm-cpus-to-2019

They might or might not catch TSMC, but Intel has a lot of money and resources. It will all come down to how well their 10nm and 7nm compare to what TSMC has. I've read at several places where people say Intel's 10nm is comparable to TSMC's 7nm production.

My personal belief (and from what I've read) is Intel will be back in full swing in 2021, and we should know for sure at that time if they are able to compete with, and/or beat TSMC's production / tech.
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
7 nm is (best case) risk risk production in 4Q 2021, and the GPGPU product being produced then is likely only for Aurora and is only in 2021 to ensure they don't lose the contract.
 

Failnaught

Member
Aug 4, 2008
26
25
91
Assuming Intel 7nm development is going well internally (ie. on track for their stated targets), should we be seeing anything from the outside yet? I guess the foundries will release things like design tools and such. Is there an equivalent for Intel?
 

moinmoin

Diamond Member
Jun 1, 2017
4,934
7,619
136
(...) but Intel has a lot of money and resources.
Both 14nm and especially 10nm have shown Intel needs time. At a time their competitors are no longer affording them that. Also the continuous polishing 14nm received backed Intel into a spec corner that 10nm fails to reach, and it's not a given 7nm will fare any better sooner.
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
Assuming Intel 7nm development is going well internally (ie. on track for their stated targets), should we be seeing anything from the outside yet? I guess the foundries will release things like design tools and such. Is there an equivalent for Intel?
Well, Intel has the same things, like the PDK, but it’s held internally.
 

ThatBuzzkiller

Golden Member
Nov 14, 2014
1,120
260
136
The telecommunication equipment advantage is mainly due to intellectual property theft on a massive scale and Huawei being owned by The Peoples Republic of China with its colossal subsidies.

That was years ago and it still doesn't explain Huawei's leadership in 5G technology as elaborated by soresu ...

That linked article showed a single Geekbench score inferior to Cortex A72 scores I have seen - even if it was a particularly low clock the GB score was running at, a high IPC at low clock does not guarantee that it will clock high enough to be competitive with Haswell, let alone Zen2.

Besides which, if it is so good then why did they even bother with the Hygon/Dhyana joint venture in the first place?

As to India, they are developing R5 cores in the Shakti program at the Madras University (IIT Madras) - I don't know where you get this strange 2 decade mumbo jumbo from, India produces a fair amount of intelligent individuals that enter the tech space, I'm fairly sure that Raja Koduri numbers among them, as are/were likely many others at AMD and Intel over the years.

Don't make the mistake of seeing the worst parts of the country in the media and judging the competence of the whole on that - there are some parts of China that are at least as backward I don't doubt.

That was just a very early engineering sample and their previous x86 series was perfectly capable of reaching 3GHz so I've no doubt that their KX-7000 series will deliver similar clock speeds on TSMC's 7nm logic node and the THATIC joint venture is only a temporary solution ...

India may produce some talented individuals but it does not mean they will necessarily contribute to India's technology sector like we see with China ...

And I would definitely argue that they're 2 decades behind technologically speaking. India just has under 3 trillion dollars in nominal GDP so it'll take at least another 2 decades to reach the same level of technological development that China has now ...

China as a whole are light years ahead of India in terms of technology and they are making real progress to contest the western technological world!

No they don't. Besides, the lion's share of chip R&D in China is already being thrown at other ISAs.

Not really, China has more money than you think and they mostly deal with very common ISAs like ARM or x86 ...
 
  • Like
Reactions: ozzy702

DrMrLordX

Lifer
Apr 27, 2000
21,583
10,785
136
Not really, China has more money than you think and they mostly deal with very common ISAs like ARM or x86 ...

Doesn't matter what resources China does or doesn't have. They aren't spending much of it in x86, period. Their biggest successful venture to date has been Huawei/HiSilicon. So long as Arm Holdings is allowed to license designs to Chinese firms, you can bet China will be pushing ARM. RISC V may or may not gain some traction, and there's still interest in MIPS designs, but x86? Certainly not. Zhaoxin is a bit player compared to the ARM behemoths of China.

edit: oops somehow a bunch of other quotes I meant to put in here got nixed. Bleh. I'll try to correct:

@UsandThem

I would love for Intel to be fully-competitive again on a 7nm EUV node in 2021. Sadly, their own roadmaps seem to indicate that 7nm will only be used for enterprise Xe until 2022. Intel has a lot riding on 10nm products like Tiger Lake, Sapphire Rapids, and Granite Rapids. It remains to be seen how well those will fare. 2022 is a long time to wait for a proper node refresh for desktop parts . . . and if the 10nm Rapids chips don't fare well, it'll be even worse.

@Adonisds

8c Skylake (let's be honest; it would be an evolution of CoffeeLake) on TSMC 7nm would be tiny and weird. Talk about heat density problems! I think it would clock about as well as Matisse.
 
Last edited:
  • Like
Reactions: Adonisds

VirtualLarry

No Lifer
Aug 25, 2001
56,229
9,990
126
8c Skylake (let's be honest; it would be an evolution of CoffeeLake) on TSMC 7nm would be tiny and weird. Talk about heat density problems! I think it would clock about as well as Matisse.
I wouldn't mind seeing an Intel Core design on TSMC fabs, 7nm would be fine. Just as sort of a limited-run test chip, of sorts. Maybe they (Intel) could put them into a new lineup of NUCs.

Would they be cheaper than Intel's current 8C Skylake-derivative CPUs, or more expensive? (Tradeoffs, smaller die on 7nm, maybe they could enlarge the L3 cache slightly), more chips/wafer, but 14nm is largely depreciated, cost-wise, I expect.
 

JasonLD

Senior member
Aug 22, 2017
485
445
136
@UsandThem

I would love for Intel to be fully-competitive again on a 7nm EUV node in 2021. Sadly, their own roadmaps seem to indicate that 7nm will only be used for enterprise Xe until 2022. Intel has a lot riding on 10nm products like Tiger Lake, Sapphire Rapids, and Granite Rapids. It remains to be seen how well those will fare. 2022 is a long time to wait for a proper node refresh for desktop parts . . . and if the 10nm Rapids chips don't fare well, it'll be even worse.

I think Granite Rapids is 7nm. Considering Intel 7nm is expected to be quite a bit denser than TSMC 5nm, if they can get desktop parts out by 2022, they should be alright.
 

soresu

Platinum Member
Dec 19, 2014
2,617
1,812
136
Not really, China has more money than you think and they mostly deal with very common ISAs like ARM or x86 ...
Mostly?

Here you are definitely being overly dramatic - neither their ARM nor x86 efforts are in any of their supercomputers as yet - unlike Shenwei's purported Alpha derivative.

Likewise their Loongson MIPS effort has been going for a decade longer than Phytium.

Their serious ARM based efforts are relatively recent - with the company being founded in 2012 (so basically ARMv8.0-A was the driving force), and with the Phytium Mars announcement coming only in 2017.

They certainly made up for that late arrival in enthusiasm though.

Zhaoxin was founded even later than Phytium in 2013 - and like Hygon required the existing VIA Isaiah design to springboard its development cycle.

This linked article demonstrates a more balanced prediction at the end despite their claimed target of matching Zen2 performance:

"Unfortunately, even with a 1.5x ST performance, they would be a fair bit behind in IPC given that the KX-5000 series appears to be slightly behind Intel’s Goldmont level of performance."
 

DrMrLordX

Lifer
Apr 27, 2000
21,583
10,785
136
I think Granite Rapids is 7nm. Considering Intel 7nm is expected to be quite a bit denser than TSMC 5nm, if they can get desktop parts out by 2022, they should be alright.

Oh yeah, Granite Rapids is 7nm. It's a 2022 chip though. If everything goes alright. Also once Intel has 7nm running for desktop/server, TSMC may be on 3nm (eek).

@soresu

Let's put things in perspective:


HiSilicon's designs pull an enormous amount of revenue each year. And most, if not all, of their CPU designs are ARM-based. Phytium is interesting, but in terms of investment/revenue, they aren't much. Regardless, if any ISA is getting major backing in China, it is ARM. HiSilicon/Huawei has their Kunpeng CPU lineup, including the 64c Kunpeng 920. They may not have made it into any supercomputers yet, but then, I do not think they much care about such things.
 
Last edited:

NostaSeronx

Diamond Member
Sep 18, 2011
3,683
1,218
136
Oh yeah, Granite Rapids is 7nm. It's a 2022 chip though.
Granite Rapids, GoldencoveX is 7nm+, so it is perfectly normal for it to be 2022.

Intel's 7nm can be done right now on existing EUV tools at Intel at LVM. They however want to use the >150 wph tools with higher uptime for production HVM.

- SRAM, ROM, Register-File: bit-memories <= These are done in the PDK.
Standard Cells and RO will be the second and should be finished within this year(2019).
Then, they will focus on eMRAM, eDRAM, and analog/eRF in 2020.
 
Last edited:

insertcarehere

Senior member
Jan 17, 2013
639
607
136
And I would definitely argue that they're 2 decades behind technologically speaking. India just has under 3 trillion dollars in nominal GDP so it'll take at least another 2 decades to reach the same level of technological development that China has now ...

China as a whole are light years ahead of India in terms of technology and they are making real progress to contest the western technological world.

And yet India, not China, has arguably the world's best telecommunications networks in terms of combining affordability, coverage, and speed.

Chinese mobile networks absolutely do not offer anywhere near 40GB of 4G data/month for $2.5 USD/month.
 

ThatBuzzkiller

Golden Member
Nov 14, 2014
1,120
260
136
Mostly?

Here you are definitely being overly dramatic - neither their ARM nor x86 efforts are in any of their supercomputers as yet - unlike Shenwei's purported Alpha derivative.

Likewise their Loongson MIPS effort has been going for a decade longer than Phytium.

Their serious ARM based efforts are relatively recent - with the company being founded in 2012 (so basically ARMv8.0-A was the driving force), and with the Phytium Mars announcement coming only in 2017.

They certainly made up for that late arrival in enthusiasm though.

Zhaoxin was founded even later than Phytium in 2013 - and like Hygon required the existing VIA Isaiah design to springboard its development cycle.

This linked article demonstrates a more balanced prediction at the end despite their claimed target of matching Zen2 performance:

"Unfortunately, even with a 1.5x ST performance, they would be a fair bit behind in IPC given that the KX-5000 series appears to be slightly behind Intel’s Goldmont level of performance."

I don't think I'm being overly dramatic all things considered ...

Your view solely being based on supercomputers is extremely narrow and the Sunway is it's own ISA and is not a derivative of the DEC Alpha ISA. Loongson's development has long since been scaled down so it's current development costs have an insignificant impact to China ...

As for ARM CPU designs, China doesn't feel an urgent need to independently develop them. They're perfectly fine with licensing from ARM Holdings who is owned by a Japanese corporation which is Softbank Group and what's more is that their current CPU designs are not of US origin so the chances of a blockade is far smaller. China as it is sees no need to react very quickly here ...

Zhaoxin despite been founded later than Phytium has seen massive progress because of a need to circumvent a US blockade on the restriction of China's importation of certain x86 products. The case with VIA is absolutely different to Hygon. THATIC doesn't design the products, they repurpose products like Hygon for a specialized market. VIA on the other hand was a small Taiwanese company that was doomed to obscurity anyways so their only viable path was tapping into chinese funds. Zhaoxin may have needed VIA's IPs to get a head start but it's clear that VIA needed the chinese money even more because it's mostly credit to Zhaoxin (who handled the vast majority of the design work since then) that VIA has gotten as far as they have with their in development KX-7000 series. In the end, THATIC was only a temporary stopgap so that Zhaoxin could have more time to bake their x86 CPU designs ...

The article you linked is also outdated, the KX-5000 is old news. The KX-7000 on the other hand is looking to be a very decent design. Haswell level IPC at 3GHz is a pretty good milestone for China ...
 

ThatBuzzkiller

Golden Member
Nov 14, 2014
1,120
260
136
And yet India, not China, has arguably the world's best telecommunications networks in terms of combining affordability, coverage, and speed.

Chinese mobile networks absolutely do not offer anywhere near 40GB of 4G data/month for $2.5 USD/month.

And yet India's mobile network operators rely heavily on equipment from Huawei Technologies Co ...

India's mobile operators were one of Huawei's biggest customers in their TD-LTE technology ...

What you are arguing is largely a matter of service and not necessarily technology. There are a multitude of reasons aside from technology why one mobile network operator will offer superior service over the others ... (Indian operators have a much lower cost of operations so it's not like any country will be able to match them in value)

India's mobile network operators may provide good service for value but it's too bad that they'll be behind the curve in terms of commercial 5G NR deployment ...
 

soresu

Platinum Member
Dec 19, 2014
2,617
1,812
136
The article you linked is also outdated, the KX-5000 is old news. The KX-7000 on the other hand is looking to be a very decent design. Haswell level IPC at 3GHz is a pretty good milestone for China ...
I believe there is a saying about counting chickens before they are hatched.

The article I linked is less than 2 years old, hardly old news - KX 5000 was only used as a reference point to the 1.5x improvement stated projection for KX 7000.
 

soresu

Platinum Member
Dec 19, 2014
2,617
1,812
136
HiSilicon's designs pull an enormous amount of revenue each year. And most, if not all, of their CPU designs are ARM-based. Phytium is interesting, but in terms of investment/revenue, they aren't much. Regardless, if any ISA is getting major backing in China, it is ARM. HiSilicon/Huawei has their Kunpeng CPU lineup, including the 64c Kunpeng 920. They may not have made it into any supercomputers yet, but then, I do not think they much care about such things.
As much as I support HiSilicon (as they seem to be the only silicon designer using both ARM's high end CPU and GPU designs regularly now), I discounted them from the conversation as they are basically using licensed core/uArch from ARM at the moment (though obviously there is homegrown tech in Kirin also).

I was discussing natively developed uArch, not whole SoC development.

I think Kunpeng 920 is one of the first examples of Neoverse N1, I will be excited to see some performance figures on that, given ARM's projected power consumption for 64 cores.

Edit: my mistake, it's not N1/A76 uArch.

It's a custom derivative of A72 called TaiShan v110 - enthusiasm dying.....
 

DrMrLordX

Lifer
Apr 27, 2000
21,583
10,785
136
As for ARM CPU designs, China doesn't feel an urgent need to independently develop them.

Uh, what?


Don't be so sure. Plus: interconnect, interconnect, interconnect. Huawei has spent a lot of time and money on CPU interconnects.

@soresu

If you discount whole SoC development, though, then you get a narrower picture of what is going on in China (and worldwide). Let's face it, most of China's domestic CPU research has been dwarfed by commercial CPU/SoC deployment research. Huawei has a lot of money. They can fund development of things like Kunpeng 920 and their custom v110 core. There's nothing else going on in China that attracts enough investment to really compare with what they are doing. Loongson/Godson has mostly flopped, RISC V is still in its infancy, and Xiaoxhin's work is more-or-less a cheap short-term alternative to China being able to ditch Wintel forever.

I think we're getting a little off-topic here, so let's bring this back around to Intel and their fab business: things would get a lot more interesting in China if/when Intel did finally drop out of the fab race. One less reason for China to keep using a foreign source for high-powered CPUs. I wouldn't expect them to try to replace Intel's x86 with a homegrown ISA, though. China is going to go for what's ubiquitous first and effective second. ARM would more-or-less fit that bill.
 
  • Like
Reactions: soresu