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

 

[511]

And Intel already sunk a ton of money into even more 18A capacity, that they will esentially just not be using because no one (including themselves) wants more 18A capacity.
Also yes, Intel ships much less volume than TSMC. This was a thing even before 18A.

lol you think you know more about Intel on how to manage fabs? also the 60WSPM is for N2 and that is shared between Apple/AMD/Qualcomm/Mediatek/Intel while the entire 20KWSPM is for Intel alone

 

[Win2012R2]

Intel will be pushing a bunch of 18A volume no matter what

Can they afford it if it loses them money per chip?

the entire 20KWSPM is for Intel alone

Yeah, that would make a lot of Celestial dGPUs

 

[Geddagod]

You can buy both now, it just took forever.

At this point, if Intel was serious about recovering their server share, they need to fab at TSMC.

I think DMR not being at TSMC is ok. Even iso node vs AMD, I don't think DMR will beat Venice Dense because one, using E-cores at that power-per core level would be more efficient than using P-cores, and two, Intel P-cores have blegh perf/watt at the lower end of the curve. Don't think PTC fixes that.
Essentially, no point in wasting a bunch of money going external, for a part that won't be on par with AMD even if it did.

 

[511]

You can buy both now, it just took forever.

cause server takes time to ramp? even at TSMC large dies take longer to ramp into volume cause you have to burn more wafers

 

[Geddagod]

lol you think you know more about Intel on how to manage fabs? also the 60WSPM is for N2 and that is shared between Apple/AMD/Qualcomm/Mediatek/Intel while the entire 20KWSPM is for Intel alone

No, I think Intel is making the right choice by going external, they are doing a great job there.
And the reason is because they need N2 to hope to be competitive, not because of any artificial capacity constrains on 18A.
Also, tbf, given how the past decade went for Intel, yes a random redditor prob could have done a better job managing the fabs LOL.

 

[511]

I think DMR not being at TSMC is ok. Even iso node vs AMD, I don't think DMR will beat Venice Dense because one, using E-cores at that power-per core level would be more efficient than using P-cores, and two, Intel P-cores have blegh perf/watt at the lower end of the curve. Don't think PTC fixes that.

PTC kind of fixes that alongside the horrible PPA of their cores they are trying that we will see lol btw they are planning Clearwwater Forest successor with AVX-512/APX and moar cores

 

[Win2012R2]

And the reason is because they need N2 to hope to be competitive, not because of any artificial capacity constrains on 18A.

They want N2 also because for every such wafer they use then AMD can't get it and in any case it will be more expensive to them.

 

[511]

Also, tbf, given how the past decade went for Intel, yes a random redditor prob could have done a better job managing the fabs LOL.

this i won't deny lol when they had Monopoly but i doubt now a redditor can do a better job

 

[511]

They want N2 also because for every such wafer they use then AMD can't get it and in any case it will be more expensive to them.

bruh this is the most horrible reason i have ever heard

 

[511]

Hopefully DMR ramps faster than GNR seemed to have.

totally depends on the chiplet die size they packing they have to do but GNR is still in ramp at Ireland

I just refuse to believe GNR allowed AMD to gain so much market share in such a short amount of time, based on their competitiveness LOL. Hopefully their Q1 market share update coming tomorrow? IIRC, would have better numbers.

AMD's Marketshare took like 6 Years from 2019 to reach 40% lol

 

[Geddagod]

Can they afford it if it loses them money per chip?

Yeah, that would make a lot of Celestial dGPUs

haha, the only way they would be burning money like that is if they allocated a bunch of 18A capacity for Celestial dgpus...
But NVL tiles seems like a good way to use 18A volume. Priority should prob go CPU, iGPU, and then IO tiles. Next up in importance is prob DMR, then CLF, then WCL close behind. Celestial should be dead last.

bruh this is the most horrible reason i have ever heard

Lowkey

btw they are planning Clearwwater Forest successor with AVX-512/APX and moar cores

Exciting

AMD's Marketshare took like 6 Years from 2019 to reach 40% lol

Mercury says they gained more server market share in Q1 2025 alone (2.1%) than they did all of 2024 (1.5%).
No way that if GNR is in good volume they lose that much market share, since GNR is decently competitive... right?
Hard to believe. Maybe this is an effect of Intel losing market stickiness in DC and less about how competitive Intel is, since they are still losing competitively anyway.

 

[511]

Mercury says they gained more server market share in Q1 2025 alone (2.1%) than they did all of 2024 (1.5%).
No way that if GNR is in good volume they lose that much market share, since GNR is decently competitive... right?
Hard to believe. Maybe this is an effect of Intel losing market stickiness in DC and less about how competitive Intel is, since they are still losing competitively anyway.

is there some recent SuperComputer order at AMD that can be due to that

 

[511]

Exciting

384 Cores iirc not sure the exact number

 

[jpiniero]

Can they afford it if it loses them money per chip?

The alternative is not getting anything for test wafers.

 

[Geddagod]

is there some recent SuperComputer order at AMD that can be due to that

Dunno. But we will see how this quarter fares, I just checked, and mercury research should be updating numbers to include Q2 today. Might be later today or maybe in a few days if it takes time for that info to be trickled down to the public.

 

[Win2012R2]

is there some recent SuperComputer order at AMD that can be due to that

Those are now heavy GPU biased, so we'd know about it, likely soon because MI400 and certainly 500 will likely reach the right level.

 

[511]

Dunno. But we will see how this quarter fares, I just checked, and mercury research should be updating numbers to include Q2 today. Might be later today or maybe in a few days if it takes time for that info to be trickled down to the public.

you have a paid license for their subscription?

 

[Geddagod]

384 Cores iirc not sure the exact number

I think how good it's gonna be is very dependent on when it comes out. Any idea about planned launch date?

 

[Geddagod]

you have a paid license for their subscription?

Oh no, they say when they will announce the results on the website, I just went and checked lol

 

[511]

Those are now heavy GPU biased, so we'd know about it, likely soon because MI400 and certainly 500 will likely reach the right level.

I have seen few AMD MI System with Xeons lol

 

[Covfefe]

Pipe cleaner for what lol

It's possible that 18A-P is different enough from 18A that it has poor yields and needs work. I'd even go so far to say that's the only plausible answer for why they'd do one single 4P 0E chip on 18A-P when 18A already had much bigger chips.

 

[Geddagod]

I have seen few AMD MI System with Xeons lol

LOL
I think Xeon may start seeing some recovery soon since they were selected to be paired with Blackwell systems. Hopefully Nvidia puts enough Intel chips in their GPU systems (though obviously they have their own CPU now too) for the server bleed to stop.

 

[Win2012R2]

The alternative is not getting anything for test wafers.

Test wafers help improve yield.

I think Xeon may start seeing some recovery soon since they were selected to be paired with Blackwell systems.

Nvidia makes a lot more money by selling their own stuff, that Xeon stuff is for paupers, chicken feed level.

 

[511]

The alternative is not getting anything for test wafers.

btw Intel paid for many test wafers for 18A during test chips

 

[jpiniero]

Test wafers help improve yield.

They are selling the test wafers. That's what Panther/Nova? is.