News Intel launches new 18A website, highlights milestones and specifications

[TerryLaze]

Except that they cannot do that and this is exactly why their stock price is ~$25 right now. As soon as they were transparent about the cost of fab operation investors punished them and wiped out around half their valuation. While I'm sure we'd both agree this is stupid it is the reality.

What does the stock price have to do with anything?!
Profits the company makes have nothing to do with the stock market.
Stock market valuation has extremely little to do with how much value a company actually has or how much money they make.

No, it was never a good business move because it was very short sighted and has cost them immense amounts of money. It also undermined future profits and allowed their competition to catch up. Had they executed 10nm (or only been ~1 year behind) then it would have paid off, but that isn't what happened.

What immense amounts of money are those, care to elaborate?!
Would building the new fabs be cheaper if intel didn't stay at 10nm for that long?!
Would intel have more money available somehow?
If it undermined anything it would be the 10nm FABs that got undermined which are pretty irrelevant by now. Or on the other hand also very mature by now for customers that could still use 10nm.

This is basic supply and demand economics where they need to be mindful to not produce so much that they lower the value of their own products.

Is that why nvidia doesn't make much money from servers? ...Oh wait!

Not selling as much as you can, when you can, just in case that you could possibly maybe hurt your margins in the long run is one way you could run your business.
As you said, they are going to sell a generation for about 2 years+ so there isn't a realistic way for them to overproduce GPUs, they would instead get all the money that is now going to scalpers, and OEMs that bundle GPUs with crap.

You seem to be ignoring the reality that the PS5 launched into and the reason behind the shortages.

What I remember are stories about how people got frustrated with waiting for ps5 to be available and went with laptops/desktops/gpus/ whatever else.
Now I know this isn't the norm by any stretch, but I am saying that that is a big reason why 'second source' is important to many big contracts.

 

[thestryker]

Is that why nvidia doesn't make much money from servers? ...Oh wait!

Enterprise is a whole other situation where we're in the middle of an AI boom and the bottleneck is packaging not silicon. This is also not even remotely comperable to the client situation since they're still producing and selling last generation's cards for that market.

Not selling as much as you can, when you can, just in case that you could possibly maybe hurt your margins in the long run is one way you could run your business.

This is absolutely how they approached the 4090.

As you said, they are going to sell a generation for about 2 years+ so there isn't a realistic way for them to overproduce GPUs, they would instead get all the money that is now going to scalpers, and OEMs that bundle GPUs with crap.

They absolutely can overproduce GPUs and end up with video cards sitting on shelves. This is a bad thing for everyone making money off of them. The only time nvidia makes money is when they sell the GPUs to AIBs. That means nvidia doesn't profit off of higher prices just product volume. However if the cards were sitting on shelves nvidia would have to cut production or lower prices to get AIBs to buy them.

What does the stock price have to do with anything?!
Profits the company makes have nothing to do with the stock market.
Stock market valuation has extremely little to do with how much value a company actually has or how much money they make.

It has everything to do with how a publicly traded company is run.

What immense amounts of money are those, care to elaborate?!

The R&D costs of developing the node and then the final cost of the node itself.

Would building the new fabs be cheaper if intel didn't stay at 10nm for that long?!

I'm not totally sure what you want to know with this question. Seeing as all of the production EUV equipment is going to existing fabs it's hard to say if there would be any literal building cost reductions.

Intel's 10nm was originally supposed to launch in 2016, but didn't end up in any volume shipping part until 2019. 10nm didn't end up covering all of Intel's parts until 2021. If Intel had bought EUV machines around the same time as TSMC they likely would have been on a similar release cadence. If I'm remembering correctly chips using TSMC's first EUV process were about the same time as ICL was launching.

If Intel had been acquiring EUV machines early on then DUV 10nm could have been less aggressive in density and likely launched when originally scheduled. They wouldn't have needed to continue to use 14nm for CPU production. They'd have also never needed to ship machines half way across the world for an accelerated ramp period.

Would intel have more money available somehow?

Yes, Intel would have had more money if they hadn't had to sink an extra 5 years into the process node.

If it undermined anything it would be the 10nm FABs that got undermined which are pretty irrelevant by now.

I don't understand what you're trying to say with this sentence.

Or on the other hand also very mature by now for customers that could still use 10nm.

Nobody but Intel can realistically use any of the 10nm process nodes because they use all internal tools. The process node is not doing Intel any favors due to the complexity and associated cost. Until the node they're designing with UMC is ready Intel's DUV fabs are largely Intel only.

 

[ManDaddio]

I hope 18A does well. But the chips (mainstream) have to be "good" and at a good price.
Gamers were a very big base for them. And they lost a good chunk since AMD went ahead.

 

[bit_user]

Companies going to intel as a 'second source' supplier to lower the danger of delays if not outright losing shipments or just to get more wafers quicker is all intel needs to at least make good business.

Porting a chip from one node to another is extremely expensive, even if they're using compatible toolchains. Tell me just how many chips have been ported between TSMC and Samsung, despite both supporting industry standard EDA toolchains for ages! Worse, the cost of doing layout is spiraling, as transistor counts continue to increase.

Companies like TSMC make a big deal about optimized nodes that are layout-compatible variants of earlier ones (e.g. N7 vs. N6), because that means the cost of porting is trivial.

 

[bit_user]

The 10nm debacle produced them tens of billions more net income for several years each.
It would have been bad if they where a foundry business back then, at least for their customers that would have wanted to move on, but for intel it was a really good business move.

No, just look at the Ice Lake Xeon. It launched too late, contributing to Intel's decline in datacenter market share. The delays of their 10 nm-family nodes probably also contributed to the lateness of Sapphire Rapids, which would've been great against the 64-core EPYCs, but struggled against the 96-core Genoa.

The 10 nm delays weren't catastrophic at the time, only because demand remained so strong and nobody else could fulfill it. However, it cast a long shadow.

You seem to be ignoring the reality that the PS5 launched into and the reason behind the shortages. Most console launches have had short term supply issues due to FOMO, but that's an expected part of the market. Sony sold a lot more consoles than they were expecting up front and their wafer buys are based on sales projections.

Not only that, but their initial wafer buys certainly anticipated they'd be able to do follow-on buys as-needed. It seems they were too slow to get in those orders for additional wafer capacity, before the fabs got backlogged by up to a year.

You seem to think that foundry capacity is the primary reason for shortages when there's no evidence this is the case outside of uncontrollable circumstances (ex: crypto boom/pandemic).

I did see a recent news story that Nvidia said it will devote more of its wafer allocation to client GPUs, due to a falloff in demand for HPC/AI Blackwells. This gives some credence to the notion that the debacle of the RTX 5000 launch was hit by lack of wafer supply, but I think it's also plausible that it was exacerbated by various bugs and quality issues that might've decimated the number of GPUs that passed all of their quality checks.

NVIDIA's GeForce RTX 50 Series Supply Might Increase Drastically In The Future; Team Green Might Dedicate AI Production Lines Towards Consumer GPUs

The supply of NVIDIA's GeForce RTX 50 GPUs may improve drastically, as Team Green might dedicate some of its AI production lines.

wccftech.com

BTW, speaking of pandemic-era GPUs, this didn't apply to Apmere, since those client GPUs used Samsung 8nm, while the A100 used TSMC N7.

 

[spongiemaster]

You can't play short term when it comes to foundry services: see the Intel 10nm debacle. The long game is the entire game, and TSMC being slightly more conservative than Samsung/Intel in recent times is why they're where they are process wise. This doesn't guarantee them victory of course, but Intel needs to be pushing to be at the top of the stack.

You have to catch up to someone before you can surpass them. There have to be attainable goals that build up to industry leader. Intel can't go from years behind to industry leader in a year or 2 and make up all the lost ground with one node. Being overly aggressive is what got Intel in the position they are in. If they had been less ambitious with 10nm, they likely wouldn't have been 4 years behind schedule getting it to market. That said, I don't see how Gelsinger's goal for Intel to release 5 nodes in 4 years doesn't qualify as "pushing."

 

[thestryker]

You have to catch up to someone before you can surpass them. There have to be attainable goals that build up to industry leader. Intel can't go from years behind to industry leader in a year or 2 and make up all the lost ground with one node. Being overly aggressive is what got Intel in the position they are in. If they had been less ambitious with 10nm, they likely wouldn't have been 4 years behind schedule getting it to market. That said, I don't see how Gelsinger's goal for Intel to release 5 nodes in 4 years doesn't qualify as "pushing."

They're not doing this in "a year or two" though it's all part of the 5N4Y plan.

If by the end of 5N4Y (7, 4, 3, 20A (canceled), 18A) Intel hasn't caught up/moved into a leadership position then it failed. You were saying all Intel has to do is beat Samsung and that just isn't a good take. Intel needs to match/beat whatever TSMC's best competing node is. That's the entire goal of the 18A node in the first place.

 

[Gaidax]

Last I heard it has like 20-30% yields. That's atrocious.

We need this up and running, but with those kinds of yields, it might as well be a fairytale.

 

[bit_user]

Last I heard it has like 20-30% yields. That's atrocious.

Yeah, I wouldn't be too surprised if Intel ended up releasing Panther Lake without backside power. That involves several additional assembly steps that could be hurting the final yield.

And without backside power, can 18A really stand up to TSMC N2?

 

[thestryker]

Last I heard it has like 20-30% yields. That's atrocious.

We need this up and running, but with those kinds of yields, it might as well be a fairytale.

Yeah I'd suggest you steer real clear of random person on the internet sharing a yield percentage without any other data. The first question you should ask when you see any "yield percentage" is what exactly is that referring to, because without that information yield percentage means literally nothing.

FYI based on that last "yield percentage" leak which was from mid 2024 PTL would be over 50% so again I go back to my first sentence.

edit: This is the closest we're actually going to get to real information any time soon due to NDAs.

(Side note here from Ian - I’ve spoken with Intel about their yield numbers. Unfortunately I can’t share exact values and refer you to Intel’s publicly available comments, but I can confirm the yield curves are following standard industry trends as we get closer to high-volume manufacturing.)

https://morethanmoore.substack.com/p/intel-2024-q4-financials

 

[Gaidax]

Yeah I'd suggest you steer real clear of random person on the internet sharing a yield percentage without any other data.

I don't really need to do this, because it literally does not matter.

Ultimately, it's down to Intel to prove viability of that node, whether one leaker or the other are right is of no consequence.

There is no point to simp for one fab or another, it's all down to what either manages to actually mass produce in the end. I'd wish Intel would be there with TSMC, but so far, they were nothing but disappointment when it comes to real manufacturing of real products from real customers in mass production.

 

[COLGeek]

Civility is required by all members. It is never okay to attack each other.

It is okay to dispute ideas with sources, facts, and other ideas.

Take it down a notch, or two, please. Thank you.

 

[Pigpig]

We are tired to see this kind of marketing strategy news already, please provide us the fact and actual result, we have seen too many news & power point slides

 

[spongiemaster]

You were saying all Intel has to do is beat Samsung and that just isn't a good take. Intel needs to match/beat whatever TSMC's best competing node is. That's the entire goal of the 18A node in the first place.

No, I did not. I said in the short term, as in 18A this year. According to you, there is no short term. Yes, there is. I don't know what your issue is in acknowledging every aim for the stars plan has smaller milestones along the way to get there. When JFK declared the goal of going to moon by a date, they didn't just go straight to the moon. There were multiple missions along the way to test procedures and equipment before the final mission to the moon in 69.

IEDM 2025 – TSMC 2nm Process Disclosure – How Does it Measure Up? - Semiwiki

Initial thoughts At IEDM held in December 2024, TSMC presented: “2nm Platform Technology featuring Energy-efficient Nanosheet Transistors and Interconnects co-optimized with 3DIC for AI, HPC and Mobile SoC Applications,” the authors are: Geoffrey Yeap, S.S. Lin, H.L. Shang, H.C. Lin, Y.C. Peng...

semiwiki.com

Based on this analysis it is our belief that Intel 18A has the highest performance for a 2nm class process with TSMC in second place and Samsung in third place.

To go from a price of <$20,000/wafer for 3nm wafers to $30,000/wafer for 2nm wafers is a >1.5x price increase for a 1.15x density improvement, that is a dramatic increase in transistor cost and it raises the question of who would pay that, our price estimates are <$30,000/wafer. There have also been reports that Apple who is typically TSMC’s lead customer for each node may be forgoing initial 2nm use due to price although we have also heard push back on that.

If TSMC prices 2nm wafers at $30,000/wafer they will create a lot of pressure for customers to switch to Intel and Samsung for 2nm class wafer supplies.

TSMC high density 2nm node is reportedly a lot denser than Intel's 18A. Though, most customers will probably be using the less dense performance variation. If Intel can deliver 18A this year with better performance and lower costs, then it is reasonable to state Intel has completed their comeback and caught TSMC.