Intel Restarts 14 nm Operations in Costa Rica, Aims to Increase Capacity for Xeon Output

[Raevenlord]

Intel has decided to restart operations in its previously winded-down Costa Rica facilities. An Intel Product Change Notification (PCN) for their Cascade Lake Xeon Scalable processors shows that the company has added Costa Rica to its three other "Test and Finish" sites - the other three are located in Penang (Malaysia), Kulim (Malaysia) and Vietnam. Intel's aim is to guarantee a "continuous supply" of the affected processors - namely, Cascade Lake second-generation Xeon Scalable processors in the Silver, Gold and Platinum lines (in both boxed and tray SKUs).

This move, which will be done in phases. The first implementation of the Costa Rica operations will be effective on April 19th, with the remaining operations to come online on August 3rd. Intel expects to reduce dependency on their other three Test and Finish sites, while being able to bolster final production capacity by some 25% with this move.



View at TechPowerUp Main Site

 

[TheLostSwede]

I bet this has nothing to do with the Wuhan virus...

 

[dont whant to set it"']

Intel's 14nm node is dead, long live Intel's 14nm node.

 

[jeremyshaw]

Slowly undoing Krzanich's mistakes one by one.

 

[gamefoo21]

My best Intel silicon almost always came from from Costa Rica.

Hopefully this gets the supply issues resolved for a bit and make more room for 10nm.

 

[TheGuruStud]

Can ya smell it? 7nm is far away.....

 

[ARF]

N7 from Intel is almost exactly 2 years away as of today:

we expect to start the 7nm period and at a with a much better profile of performance over that starting at the end of 21

Intel: 10nm Is Going To Be Less Productive Than 22nm, 10+ Will Be Pushed Out This Year

Intel's CFO George Davis presented at the Morgan Stanley's TMT conference and had a refreshingly honest and candid discussion with analysts about the 10nm process. The company revealed that it's planning to push out 10+ process this year but also warned that while it is almost certainly in the...

wccftech.com

 

[TheGuruStud]

N7 from Intel is almost exactly 2 years away as of today:

Intel: 10nm Is Going To Be Less Productive Than 22nm, 10+ Will Be Pushed Out This Year

Intel's CFO George Davis presented at the Morgan Stanley's TMT conference and had a refreshingly honest and candid discussion with analysts about the 10nm process. The company revealed that it's planning to push out 10+ process this year but also warned that while it is almost certainly in the...

wccftech.com

So expect actual volume sometime 2023 lol

 

[ARF]

So expect actual volume sometime 2023 lol

Yup, it's bizarre. But let's give them at least some credit for the fact that they have been ambitious and their N7 node is everyone else's N5.
And their N10 node is everyone else's N7.

 

[Octopuss]

Someone explain why would a company abandon a "factory" that likely cost billions of dollars to build only to rebuild it in Asia (at least that's the impression I get out of reading this and the linked article).

 

[thesmokingman]

Can ya smell it? 7nm is far away.....

The irony is Intel keeps saying they will have parity in 2021. Its the same thing all over again, their shit don't smell to them at least.

 

[TheGuruStud]

Yup, it's bizarre. But let's give them at least some credit for the fact that they have been ambitious and their N7 node is everyone else's N5.
And their N10 node is everyone else's N7.

But too bad TSMC will be on their 3nm, and unlike the dumb gorilla, their roadmaps aren't fantasy.

Oh, remember all the talk of their new stuff being backport capable? They have ZERO confidence in manufacturing.

 

[Vya Domus]

Yup, it's bizarre. But let's give them at least some credit for the fact that they have been ambitious and their N7 node is everyone else's N5.
And their N10 node is everyone else's N7.

But that's exactly the same reason 10nm turned out so bad, they are once again trying to aim for something that's unrealistic.

 

[dicktracy]

Selling everything they make must feel so good.

 

[Dave65]

Mommy, Jason is on 3nm and I am still on 14nm.. BOO HOO HOO..

 

[newtekie1]

Someone explain why would a company abandon a "factory" that likely cost billions of dollars to build only to rebuild it in Asia (at least that's the impression I get out of reading this and the linked article).

If you read between the lines of the 2014 article, when they shut down the Costa Rica factory, it basically boils down to Intel asked Costa Rica for a bunch of tax breaks and Costa Rica didn't give them what they wanted. So it was cheaper to shut down the factory in Costa Rica and open new factories in Asia. The Asian governments likely gave them huge tax incentives to set up shop there.

 

[RandallFlagg]

Yup, it's bizarre. But let's give them at least some credit for the fact that they have been ambitious and their N7 node is everyone else's N5.
And their N10 node is everyone else's N7.

That's not really correct. Yes, Intel's 14nm is probably as good as everyone else's 12nm. That speech Intel gave about the nomenclature in 2018 which you're obliquely referring to was half-truths, typical tactic of a politician and marketer. I don't think Intel's 10nm will be as good as the first gen 7nm that has been online since late 2018 based on initial tests (yes, there are a few of gen 10 out now), and definitely not as good as the 7nm+ node that is online now.

Throw in that 5nm is slated for Q2 2020 mass production, and even if we make some adjustments for nomenclature and assume that Intel will be able to ramp up 10nm (that is probably a bad assumption given that they are bringing more 14nm back online), and their best case is that they will be 2 entire process node generations behind TSMC and friends if they have 10nm in full swing by 2H 2020.

Intels fail here is epic. They have been on 14nm since 2014. The only reason they have been able to fail so badly and keep market share is because of the work that was done to get them to 14nm far before their competition. But that work was done 10 years ago, and I suspect those folks aren't there anymore.

Changing a CEO and making it to one more node this year is not going to fix that. It is probably a corporate structure / corporate culture problem, one of those places where perception takes precedence over reality. It's been my experience that perception can win for a while, but in the end reality kicks perception in the rear. This type of long term failure is the result.

 

[hat]

They need a better architecture more than they need a smaller node, though they do need that too. It's looking very much like Pentium vs Athlon these days.

 

[dicktracy]

They need a better architecture more than they need a smaller node, though they do need that too. It's looking very much like Pentium vs Athlon these days.

They already have better architectures than AMD. Intel is continuing their development of nextgen CPU archs like Golden Cove regardless of fabs (Icelake and Tigerlake designs are done ages ago and are all superior to Zen2 and most likely Zen3 as well). Golden Cove design is expected in 2021 which is another level above AMD. Only thing preventing them from releasing those behemoths is the fabs.

All in all, fabs and CPU design teams are independent of one another.

 

[hat]

Vaporware doesn't count. I also don't care about an architecture that might have better IPC if it can't clock high enough to matter.

 

[ARF]

That's not really correct. Yes, Intel's 14nm is probably as good as everyone else's 12nm. That speech Intel gave about the nomenclature in 2018 which you're obliquely referring to was half-truths, typical tactic of a politician and marketer. I don't think Intel's 10nm will be as good as the first gen 7nm that has been online since late 2018 based on initial tests (yes, there are a few of gen 10 out now), and definitely not as good as the 7nm+ node that is online now.

Throw in that 5nm is slated for Q2 2020 mass production, and even if we make some adjustments for nomenclature and assume that Intel will be able to ramp up 10nm (that is probably a bad assumption given that they are bringing more 14nm back online), and their best case is that they will be 2 entire process node generations behind TSMC and friends if they have 10nm in full swing by 2H 2020.

Intels fail here is epic. They have been on 14nm since 2014. The only reason they have been able to fail so badly and keep market share is because of the work that was done to get them to 14nm far before their competition. But that work was done 10 years ago, and I suspect those folks aren't there anymore.

Changing a CEO and making it to one more node this year is not going to fix that. It is probably a corporate structure / corporate culture problem, one of those places where perception takes precedence over reality. It's been my experience that perception can win for a while, but in the end reality kicks perception in the rear. This type of long term failure is the result.

Intel's N10 is 2.5 times denser than its N14! And as dense as TSMC's N7.

7nm vs 10nm vs 14nm: Fabrication Process - Tech Centurion

Everything about 7nm, 10nm and 14nm Fabrication Process of TSMC, Intel, and Samsung. Transistor Density & Ranking of different technology nodes.

www.techcenturion.com

 

[yeeeeman]

N7 from Intel is almost exactly 2 years away as of today:

Intel: 10nm Is Going To Be Less Productive Than 22nm, 10+ Will Be Pushed Out This Year

Intel's CFO George Davis presented at the Morgan Stanley's TMT conference and had a refreshingly honest and candid discussion with analysts about the 10nm process. The company revealed that it's planning to push out 10+ process this year but also warned that while it is almost certainly in the...

wccftech.com

N7 from Intel is equivalent to N5 from TSMC...so they are say 1 year behind.

They already have better architectures than AMD. Intel is continuing their development of nextgen CPU archs like Golden Cove regardless of fabs (Icelake and Tigerlake designs are done ages ago and are all superior to Zen2 and most likely Zen3 as well). Golden Cove design is expected in 2021 which is another level above AMD. Only thing preventing them from releasing those behemoths is the fabs.

All in all, fabs and CPU design teams are independent of one another.

I agree and this shows just how badly informed people are.
Skylake uArch is just a tad slower than Zen 2. Yes, read it again. It has several advantages over Zen 2, like better latency, AVX 512 and some disadvantages like lower multi core scaling, 5% lower IPC. Add the same amount of cache to Skylake like in Zen 2 products and you will probably get the same IPC.
Now, Intel has other uArchs in the pipeline, you can imagine that. Ice Lake was supposed to be launched in 2017 and compete with Zen 1. Ice lake would have something like 30-35% higher IPC. Imagine that! Then Tigerlake in 2018. Then Alder Lake in 2019. And Meteor Lake in current days. So it is not like Intel doesn't have the uArch to beat AMD. They have better IP, but THEY DON"T HAVE THE MEANS TO FAB IT. Very hard to understand for some people..............
So in other words, if Intel wouldn't have had problems with fabrication, AMD would be much less successful today. Zen uArch is nothing special. It uses the same concepts Intel introduced with Sandy Bridge and its only merit is the chiplet architecture which allowed AMD to scale the number of cores to crazy amounts.

 

[ARF]

If they have so large IPC lead, why did they go so aggressively with the node ?

They wanted to bankrupt AMD once and for all ?

BTW, this is serious, if Intel's N10 had been successful in 2017, nowadays maybe there wouldn't have been any AMD left today!



edit: This is not a good question. We know that high IPC is a result of short pipeline and goes with lower frequencies.
And in order to keep the frequencies up a bit, they tried to develop a very aggressive next-gen node, which failed.

They do have 35% IPC lead but at 3.2 GHz and only a quad-core design, it won't work!

 

[ratirt]

If they have so large IPC lead, why did they go so aggressively with the node ?

They wanted to bankrupt AMD once and for all ?

BTW, this is serious, if Intel's N10 had been successful in 2017, nowadays maybe there wouldn't have been any AMD left today!



edit: This is not a good question. We know that high IPC is a result of short pipeline and goes with lower frequencies.
And in order to keep the frequencies up a bit, they tried to develop a very aggressive next-gen node, which failed.

They do have 35% IPC lead but at 3.2 GHz and only a quad-core design, it won't work!

Do they have 35% IPC lead? Sorry but I have never seen that. All your claims with this LAKEs' are futile. It all looks good on a paper an everything when Intel provides information how far advanced they arch is in comparison to others. Then the CPU actually comes out and you find out it is not 35% but merely 4% and most of this gain is due to frequency bump.
I'm sorry, but I'm not convinced. I'm not gonna hold my breath for this. When the CPUs are released, benchmarked, we will see what they actually bring to the table.

 

[ARF]

They have got at least the claimed 18% IPC increase:

Intel launches 10nm Ice Lake processors. They have 18% better IPC

vaaju.com

Intel’s Confusing Messaging: Is Comet Lake Better Than Ice Lake?

www.anandtech.com