Friday, March 14th 2025

Samsung Foundry in Trouble, Might Cancel 1.4 nm Node High-Volume Manufacturing

Samsung Foundry could abandon its 1.4 nm (SF1.4) process node initially targeted for 2027 production, according to industry leaker @Jukanlosreve. This decision comes amid ongoing yield problems with the company's 3 nm SF3 node and follows the shutdown of underutilized 5 nm and 7 nm production lines. This could significantly impact Samsung's technology roadmap, which had positioned SF1.4 alongside automotive-focused SF2A and SF2Z nodes. According to Korea Economic Daily data, the foundry division's market share has fallen to 8.2% compared to TSMC's 67.1%. Samsung continues the development of Exynos 2600 on the SF2 node and maintains orders from Japan's Preferred Networks but has failed to attract major clients beyond Chinese firms, avoiding US sanctions. Internal restructuring appears imminent, with reports suggesting the Exynos design team may move under the Mobile Experience division.

While Samsung may prioritize improving yields on existing nodes rather than pursuing SF1.4, this approach risks competitive disadvantages in high-performance computing and AI markets. The SF2Z node with Backside Power Delivery Network (BPDN) technology remains under development, though its commercial viability depends on resolving broader manufacturing issues. Samsung's retreat would further fall behind advanced node manufacturing, competing with TSMC and potentially Intel, raising questions about competition in leading-edge semiconductor fabrication. The company's decisions in the coming months will determine whether it can regain manufacturing credibility or face production of trailing-edge semiconductor nodes. Manufacturing advanced silicon remains a challenge for everyone except TSMC.
Sources: @Jukanlosreve on X, via Notebookcheck
Add your own comment

39 Comments on Samsung Foundry in Trouble, Might Cancel 1.4 nm Node High-Volume Manufacturing

#26
Dr. Dro
ScaLibBDP>>...according to industry leaker @Jukanlosreve...

Wow! Just Wow! Who is that Magician Guy? I mean @Jukanlosreve...

Let me put it in a different way. Techpowerup is an absolute spreader of tech-rumours and, very often, tech-bullshit-like news!

The red line is crossed and this is my last post. I'll Never visit Techpowerup again.
I mean, tech news is largely based on rumor and it seems to get the ball rolling. Although, I think rumor posts should be tagged as such and filed under a different subsection.

Instead of being super angry and outraged, let's pitch this idea to W1zzard.
AthenaUgh, this means we are going to be supply constrained for pretty much every product launch for the foreseeable future

hopefully they will fix the yield issue(s)
Yeeep. I don't expect the supply situation on high-performance silicon to improve any time soon. That's why I jumped at the chance to get a "MSRP" 5090... even though it's taking forever for the order to be fulfilled. It'll get comfy to bunk down and wait it out. I don't see the low prices on GPUs we saw from 2017-2019 ever returning though.
Posted on Reply
#27
TheinsanegamerN
kondamin3 year old "node" but yeah, those chips have no business being on the most in demand node.
Mainstream graphics could well be made on 7nm.
being on 7nm would give you a major disadvantage in power draw. After one generation, you'd either have to engineer new midrange parts for older nodes while simultaneously designing the same arch for newer nodes (which makes it significantly more expensive) or sell high end cards cheaper as midrange (unfeasable with current wafer prices).
kondamin3 Graphics card vendors isn't enough.
The market can barely sustain 2. If it wasnt for CPUs intel would be bankrupt, and AMD can only manage a 10% marketshare.
sephiroth117Nintendo could have used 5 or 7nm lol for the switch 2 if there was no strong demand, apparently they are on 8nm for it...
I suppose Nvidia could not offer something affordable enough besides there 8nm rumored T239 anyways
So they could pay 2-3x as much per wafer for some tech nerds to drool over? 99% of console buyers dont know or care what a process node is. Nobody ahs bought nintendo for processing power since the days of the n64 anyway.
QuicksExynos is horseshit anyways.
Strange that they don't have buyers for 5nm & 7nm. Surely many don't need a smaller process then that.

Hopefully the get their act together because having TSMC supply basically the entire world is never a good idea.
If you dont need significant compute power, why would you pay for 5 or 7nmw hen you can pay 10% as much for a 28nm chip that performs just as well?

It's a very narrow niche that sees a significant benefit from smaller nodes, can justify the price, but doesnt need cutting edge.
Dr. DroYeeep. I don't expect the supply situation on high-performance silicon to improve any time soon. That's why I jumped at the chance to get a "MSRP" 5090... even though it's taking forever for the order to be fulfilled. It'll get comfy to bunk down and wait it out. I don't see the low prices on GPUs we saw from 2017-2019 ever returning though.
We just had over half a year of GPUs like the 7900xt/xtx and 7800xt being available under MSRP, significantly so for the 7900 series. I saw 7900xtx cards going for $820 and 7900xts going for $650.

This exact same thing happened with the 3000 launch, everyone swore we'd NEVER have GPUs ins tock again and MSRP would never be reached. Give it 6 months and stock will be regularly available, give it a year and a half and I'd bet that you can find some cards under MSRP.

People have short memories. I remember people whining about the 1080ti being $700 when the 680 and 580 were $500 cards, and bemoaning the death of the high end card at the hands of nGreedia. Oh how times change.....
Posted on Reply
#28
Wirko
Dr. DroAlthough, I think rumor posts should be tagged as such and filed under a different subsection.
Yes. Wccftech does that.
Posted on Reply
#29
kondamin
TheinsanegamerNThe market can barely sustain 2. If it wasnt for CPUs intel would be bankrupt, and AMD can only manage a 10% marketshare.
nothing the market, the market turned in to a monopoly when amd took ati over and did jack all with the ip for over a decade.
Posted on Reply
#30
Dr. Dro
TheinsanegamerNWe just had over half a year of GPUs like the 7900xt/xtx and 7800xt being available under MSRP, significantly so for the 7900 series. I saw 7900xtx cards going for $820 and 7900xts going for $650.

This exact same thing happened with the 3000 launch, everyone swore we'd NEVER have GPUs ins tock again and MSRP would never be reached. Give it 6 months and stock will be regularly available, give it a year and a half and I'd bet that you can find some cards under MSRP.

People have short memories. I remember people whining about the 1080ti being $700 when the 680 and 580 were $500 cards, and bemoaning the death of the high end card at the hands of nGreedia. Oh how times change.....
That was before the latest AI boom that DeepSeek caused, though. Before that, the RDNA 3 cards weren't selling very well to gamers, and a lot of people were anticipating Blackwell to have a smooth and well executed launch (LMAO), so why pay the prices they were asking for the 40 series? If we only knew. :nutkick:
Posted on Reply
#31
Assimilator
All the morons in this thread saying "just use an older node" gee you don't think the people designing these chips have thought of this? And that maybe, just maybe, there's a reason they can't use those nodes?
Posted on Reply
#32
kondamin
AssimilatorAll the morons in this thread saying "just use an older node" gee you don't think the people designing these chips have thought of this? And that maybe, just maybe, there's a reason they can't use those nodes?
Sure scale, scale is the main reason.
but that becomes a Non issue when we are talking about the lower end skus

heck Nvidia would still be selling higher end 30 series cards if they were still making them.
Posted on Reply
#33
SRB151
kondaminnothing the market, the market turned in to a monopoly when amd took ati over and did jack all with the ip for over a decade.T
That's not completely true, AMD was competitive for quite a while, hardware wise. What bit AMD hard was SOFTWARE. Nvidia had CUDA all to themselves, and was always looking for an angle that had ZIP to do with improving the GPU side substantially. Physx, CUDA, RTX, DLSS (upscaling), and now frame generation. Nvidia has always wanted a proprietary edge to corner the market, and their raster improvements have been underwhelming for a while. And when you have to blind develop a new tech from scratch to avoid patent infringement, you start out way behind to begin with. Even then, when Nvidia gets their hooks into game developers, the games are written in such a way that it takes AMD forever to get cloase, and then it's not enough. Indiana Jones is a great example. the RX 9070 XT looks more like the 7000 serise in that title due to the new "path tracing" that changed the entire ray tracing game.

50570 is as fast at the 4090? Not hardly, but it has all the software toys. Only thing saving AMD wiith this card was the idiotic decision to ship it with 12gb.

Cant wait to see what "improvements" that Nivida and Nvidia with have next round to prevent AMD and Intel from working......especially since game develpers are now making Nvidia features a REQUIREMENT to play the game.
Posted on Reply
#34
Nhonho
Thus, Samsung loses highly profitable customers like Nvidia.
Posted on Reply
#35
x4it3n
JermelescuThe GeForce 5070 is on cutting edge tech for nothing, you have older and arguably better stuff that do the same thing, except a few perks.
Cutting edge technology? Uhhh TSMC N3P or N3X would have been cutting edge but Nvidia are still using a TSMC 4nm node...

Blackwell is pretty much a Refresh of Lovelace, the only real architectural difference (as far as I know) is that on Blackwell all the CUDA Cores can do FP32 or INT32, whereas on Lovelace and Ampere only half of the cores can do FP32 or INT32, so it can be beneficial when using all cores as FP32 or all cores as INT32, but in Gaming about 30% of the cores are being used for INT32 so it doesn't change anything.

Lovelace was a much bigger gap than Ampere because the CUDA Cores count went up by 56% vs 3090 (non Ti), it had much higher Clocks (2520MHz vs 1695Mhz) and the L2 Cache increased a lot too (72MB vs 6MB), while using a much better node (TSMC 4N vs Samsung 8nm).

The 5090 has ~33% more CUDA Cores than the 4090, a 512-bit Memory Bus with GDDR7 Memory (~78% more Memory Bandwidth) and 96MB or L2 Cache vs 72MB for the 4090 (even though both are gimped compared to a full GB202 or AD102 chip), but except that, the IPC hasn't changed, the Core clocks are almost the same, and it uses a lot more power too!
If the 5090 had been made on a TSMC 3nm node then I'm sure the 5090 would be at least 40-50% more powerful than the 4090 (if not more).

Blackwell is NOT a "new architecture" with much better IPC, etc., It just has better Tensor Cores (AI cores) and that's what Nvidia are trying to sell as "performance". FG/MFG ≠ real FPS, the Input Lag is not the same and the generated frames can have a lot of artifacts (sure it will probably get better over time, but it's not the same as of now).

PS: I'm sure Blackwell would have been much better if AMD had not cancelled NAVI 41.
Posted on Reply
#36
Jermelescu
x4it3nCutting edge technology? Uhhh TSMC N3P or N3X would have been cutting edge but Nvidia are still using a TSMC 4nm node...

Blackwell is pretty much a Refresh of Lovelace, the only real architectural difference (as far as I know) is that on Blackwell all the CUDA Cores can do FP32 or INT32, whereas on Lovelace and Ampere only half of the cores can do FP32 or INT32, so it can be beneficial when using all cores as FP32 or all cores as INT32, but in Gaming about 30% of the cores are being used for INT32 so it doesn't change anything.

Lovelace was a much bigger gap than Ampere because the CUDA Cores count went up by 56% vs 3090 (non Ti), it had much higher Clocks (2520MHz vs 1695Mhz) and the L2 Cache increased a lot too (72MB vs 6MB), while using a much better node (TSMC 4N vs Samsung 8nm).

The 5090 has ~33% more CUDA Cores than the 4090, a 512-bit Memory Bus with GDDR7 Memory (~78% more Memory Bandwidth) and 96MB or L2 Cache vs 72MB for the 4090 (even though both are gimped compared to a full GB202 or AD102 chip), but except that, the IPC hasn't changed, the Core clocks are almost the same, and it uses a lot more power too!
If the 5090 had been made on a TSMC 3nm node then I'm sure the 5090 would be at least 40-50% more powerful than the 4090 (if not more).

Blackwell is NOT a "new architecture" with much better IPC, etc., It just has better Tensor Cores (AI cores) and that's what Nvidia are trying to sell as "performance". FG/MFG ≠ real FPS, the Input Lag is not the same and the generated frames can have a lot of artifacts (sure it will probably get better over time, but it's not the same as of now).

PS: I'm sure Blackwell would have been much better if AMD had not cancelled NAVI 41.
I'm just saying you don't need the latest tech for anything in 99% of the cases. Latest mobile phones, manufacturing node, gddr7 and so on.
Posted on Reply
#37
x4it3n
JermelescuI'm just saying you don't need the latest tech for anything in 99% of the cases. Latest mobile phones, manufacturing node, gddr7 and so on.
Yes for Smartphones and Handheld consoles but No for PC Gaming since modern games are becoming very demanding, mostly games running on UE5 games and/or games using heavy RT or PT (which will be more and more common in Next-Gen games).
Posted on Reply
#38
Jermelescu
x4it3nYes for Smartphones and Handheld consoles but No for PC Gaming since modern games are becoming very demanding, mostly games running on UE5 games and/or games using heavy RT or PT (which will be more and more common in Next-Gen games).
In PC world 95%+ of the people can't afford the newest and fastest thing, so it doesn't matter.
Posted on Reply
#39
x4it3n
JermelescuIn PC world 95%+ of the people can't afford the newest and fastest thing, so it doesn't matter.
Yes but that's how you push technology... without high-end GPUs you don't push the envelope really far... Except the 4090 and 5090 there is no GPU that can run PT at a "playable" framerate. And even the 4090 and 5090 need DLSS + FG...
Posted on Reply
Add your own comment
Nov 22nd, 2026 16:11 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts