Discussion Leading Edge Foundry Node advances (TSMC, Samsung Foundry, Intel) - [2020 - 2025]

[DisEnchantment]

TSMC's N7 EUV is now in its second year of production and N5 is contributing to revenue for TSMC this quarter. N3 is scheduled for 2022 and I believe they have a good chance to reach that target.


N7 performance is more or less understood.


This year and next year TSMC is mainly increasing capacity to meet demands.

AnandTech Forums: Technology, Hardware, Software, and Deals

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

www.anandtech.com

For Samsung the nodes are basically the same from 7LPP to 4 LPE, they just add incremental scaling boosters while the bulk of the tech is the same.

Samsung is already shipping 7LPP and will ship 6LPP in H2. Hopefully they fix any issues if at all.
They have two more intermediate nodes in between before going to 3GAE, most likely 5LPE will ship next year but for 4LPE it will probably be back to back with 3GAA since 3GAA is a parallel development with 7LPP enhancements.

AnandTech Forums: Technology, Hardware, Software, and Deals

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

www.anandtech.com

Samsung's 3GAA will go for HVM in 2022 most likely, similar timeframe to TSMC's N3.
There are major differences in how the transistor will be fabricated due to the GAA but density for sure Samsung will be behind N3.
But there might be advantages for Samsung with regards to power and performance, so it may be better suited for some applications.
But for now we don't know how much of this is true and we can only rely on the marketing material.

This year there should be a lot more available wafers due to lack of demand from Smartphone vendors and increased capacity from TSMC and Samsung.
Lots of SoCs which dont need to be top end will be fabbed with N7 or 7LPP/6LPP instead of N5, so there will be lots of wafers around.

Most of the current 7nm designs are far from the advertized density from TSMC and Samsung. There is still potential for density increase compared to currently shipping products.
N5 is going to be the leading foundry node for the next couple of years.

For a lot of fabless companies out there, the processes and capacity available are quite good.

---------------------------------------------------------------------------------------------------------------------------------------------------


FEEL FREE TO CREATE A NEW THREAD FOR 2025+ OUTLOOK, I WILL LINK IT HERE

 

[The Hardcard]

50% more per wafer (not final MSPR).

It's not greedy for AMD to have good margins, they need it, but I don't want subpar stuff - 75mm chip that costs double on 2nm is still very cheap, enough consumers will accept $100 extra for really top of the range stuff - very highly clocked with lots of extra 3D cache, stuff like that.

At 75mm AMD should be getting 700+ chiplets per wafer - that's $30 at $20000 wafer price! Yes, sure they will go to servers - but put some into consumer space, sell it for $1k as best CPU available!

How are you figuring $20,000 for N2 when current wafers are around $20,000. Expect N2 to push $28,000 if not higher. Then make sure you are factoring in 15 to 30 percent defects in your math.

You won’t be getting N2 chips cheap, unless you wait until 2030 to buy.

EDIT: Given that there’s little point in using leading nodes unless you are going to be technologically aggressive across the package, you would probably be looking at 50 percent to 65 percent higher prices for AMD to force its way in to be an early customer.

Qualcomm purportedly wants $240 for a 120mm2 thereabouts N3E SOC.

 

[Meteor Late]

Same was said about N3 and Apple is maintaining prices just fine over N5.

 

[The Hardcard]

Same was said about N3 and Apple is maintaining prices just fine over N5.

Apple maintains base prices by relying on sky high upgrade pricing. No one else is doing that. You want $200 for an extra 16 GBs of RAM and another $200 for 256 GB of storage to industry wide?

EDIT: Apple is able to prepay to get TSMC to the next node because part of buying an Apple device is prepaying for the next node.

 

[Meteor Late]

Apple maintains base prices by relying on sky high upgrade pricing. No one else is doing that. You want $200 for an extra 16 GBs of RAM and another $200 for 256 GB of storage to industry wide?

Yes, I much prefer that, not with 8GB base like in the past, but with 16GB base, absolutely, and having the possibility to use external storage. Let stupid enthusiast people pay for the "Apple tax upgrade" so that value conscious customers can still get nice chips.
Besides, AMD is charging obscene amounts of money for some 4nm parts like Strix Point regardless.

 

[Win2012R2]

How are you figuring $20,000 for N2 when current wafers are around $20,000. Expect N2 to push $28,000 if not higher. Then make sure you are factoring in 15 to 30 percent defects in your math.

Sorry, my bad - yes, at $30k per wafer of N2 it will cost AMD $45 for 75mm chiplet instead of $30, I am happy for them to DOUBLE their usual margins and charge extra $30, heck, I'd pay $100 premium for that and lots of people also will. AMD should be doing it for premium high freq EPYCs and throw some crumbs into retail market inform of Legendary RyZens - sell them for $1000 if necessary, plenty of people will pay the premium to pair best CPU with future 6090s.

Qualcomm purportedly wants $240 for a 120mm2 thereabouts N3E SOC.

That's a full SOC with lots of patented by them tech (assuming this is mobile modem stuff), AMD does not have that extra cost in Zen chiplets that are almost half the size - it's totally doable if AMD gets winning mindset - premium products FIRST! Leave Intel the Celeron market.

 

[adroc_thurston]

AMD should stop mucking around and give CPUs the best process treatment

They did, you can buy Turin-D just fine.

 

[Win2012R2]

They did, you can buy Turin-D just fine.

I was talking about consumers, in Turin-D clocks too low for the intended purpose - gaming with 5090 etc.

 

[adroc_thurston]

I was talking about consumers

Consumers don't have the money to afford N3e at comfortable for AMD margins.
Better luck next time!

 

[Win2012R2]

Consumers don't have the money to afford N3e at comfortable for AMD margins.

4090 sold really well for 2 grand and 5090 will sell too, paying extra couple of hundred bucks for 75mm uber fast chip is a no brainer for most of such owners - 8 cores will do fine (12 better tho), the only way to properly increase CPU sell price is to offer real premium tech - otherwise it's race to the bottom.

 

[adroc_thurston]

the only way to properly increase CPU sell price is to offer real premium tech

No, you do this by eliminating comp.

 

[jpiniero]

Consumers don't have the money to afford N3e at comfortable for AMD margins.
Better luck next time!

Apple can make it work. Course PC consumers don't pay Apple prices for laptops.

 

[Win2012R2]

No, you do this by eliminating comp.

There is money on the table right now, good billion of revenue, NVIDIA would have never let this opportunity slide which is why they are so ahead now.

 

[Hulk]

4090 sold really well for 2 grand and 5090 will sell too, paying extra couple of hundred bucks for 75mm uber fast chip is a no brainer for most of such owners - 8 cores will do fine (12 better tho), the only way to properly increase CPU sell price is to offer real premium tech - otherwise it's race to the bottom.

What are the sales numbers for the 4090? I'm curious as to how many people outside of reviewers and competitive gamers have $2000+ to spend on a GPU. Not that I don't doubt it but am curious to what "sold really well" means?

 

[Win2012R2]

What are the sales numbers for the 4090?

It's nearly 1% on Steam Survey, which is known to overcount cheaper cards due to them being shared by players in Asian cafes. Steam is estimated to have more than 100 mln users, so we are looking here at 1 mln+ 4090s.

More were bought by pros for AI workloads too, it's a very successful top tier consumer card because for the money it offered very large perf delta, I reckon more than 2 mln cards were sold - margin on those should be excellent also.

I bought a new CPU (7950X3D) mainly to keep 4090 occupied in 4K res, and we've got 5090 coming just in a month - this will make new 9800X3D/9950X3D a must have for those configs.

 

[Markfw]

What are the sales numbers for the 4090? I'm curious as to how many people outside of reviewers and competitive gamers have $2000+ to spend on a GPU. Not that I don't doubt it but am curious to what "sold really well" means?

I have 5 and many of my fellow DC'ers have quite a few, who are not gamers.

 

[Hulk]

It's nearly 1% on Steam Survey, which is known to overcount cheaper cards due to them being shared by players in Asian cafes. Steam is estimated to have more than 100 mln users, so we are looking here at 1 mln+ 4090s.

More were bought by pros for AI workloads too, it's a very successful top tier consumer card because for the money it offered very large perf delta, I reckon more than 2 mln cards were sold - margin on those should be excellent also.

I bought a new CPU (7950X3D) mainly to keep 4090 occupied in 4K res, and we've got 5090 coming just in a month - this will make new 9800X3D/9950X3D a must have for those configs.

So we're looking at 1 million 4090's? Roughly. That $2 billion US dollars. Based on 100 million GPU's "out there" and 1% of them being 4090's, I wonder what percentage of the total sales dollars is 4090's? 5 or 6% maybe? Not bad market penetration for a halo product.

 

[Win2012R2]

So we're looking at 1 million 4090's? Roughly.

That's minimum. More importantly margin on it far bigger than 4060 and once next gen is released 4090 owners won't be flooding the market with cheap card that competes with new low end, this is all big money for NVIDIA.

5090 will do even better if they really give at least 60-70% perf and keep price under $2500 - each of those cards will need a top CPU as 8k rendering isn't a thing and 4k can get pretty high frame rates even now.

 

[Meteor Late]

When you think about it, Nvidia is reverse Apple.
Apple gives you a good value with the base model and shit value with the upgrades. Whereas Nvidia gives you a good value with the top of the line card and shit value at the lower stack.

 

[QuickyDuck]

When you think about it, Nvidia is reverse Apple.
Apple gives you a good value with the base model and shit value with the upgrades. Whereas Nvidia gives you a good value with the top of the line card and shit value at the lower stack.

The more you buy, the more you save

 

[Hulk]

When you think about it, Nvidia is reverse Apple.
Apple gives you a good value with the base model and shit value with the upgrades. Whereas Nvidia gives you a good value with the top of the line card and shit value at the lower stack.

What am I missing on the value of the 4090? According to TechPowerUp's latest pricing the 4090 is just about the worst in terms of performance for the dollar?

ASRock Arc B580 Steel Legend Review

ASRock’s Arc B580 stands out with its impressive triple-slot, triple-fan cooling solution that achieves whisper-quiet noise levels. The stylish white design enhances its appearance, making it a perfect option for white builds using an affordable GPU.

www.techpowerup.com

 

[igor_kavinski]

What am I missing on the value of the 4090?

I think it would look different in a cost per raytraced frame or cost per DLSSed frame chart.

 

[Markfw]

I think it would look different in a cost per raytraced frame or cost per DLSSed frame chart.

I also forgot to mention servers. In the DC community, a lot of people rent server power, meaning servers with 4090s. We are talking a LOT of 4090s in the cloud. Its not all about gaming or raytracing. I can't show you any numbers, but most likely in the millions of servers. Many of them have 4-8 4090s EACH server. Its simply the most powerful card/device out there.

Comment from discord:

there's a host with 8x4090's for $2.564/hr

 

[Meteor Late]

What am I missing on the value of the 4090? According to TechPowerUp's latest pricing the 4090 is just about the worst in terms of performance for the dollar?

ASRock Arc B580 Steel Legend Review

ASRock’s Arc B580 stands out with its impressive triple-slot, triple-fan cooling solution that achieves whisper-quiet noise levels. The stylish white design enhances its appearance, making it a perfect option for white builds using an affordable GPU.

www.techpowerup.com

Yeah but at least you get a very good upgrade over the last generation equivalent, not very compromised cards, I mean, even ignoring the 8GB cards, you have cards like RTX 4060 ti 16GB which sell for about 500€ in europe with only a 128 bit bus, what a joke, whereas in the past, a 60 class GPU was always a 192 bit bus with a decent cuda core count.

 

[511]

Why do i discover good thread topic so late 🤣

 

[LightningZ71]

No kidding! This place will go ->POOF<- any day now and people still haven't gotten a consensus on where we're all going.

Oh, and Foundry nodes rock! (to keep on topic)