News Gelsinger fires back at recent stories about 18A's poor yields, schools social media commenters on defect densities and yields

[Admin]

Ex-Intel CEO Gelsinger is 'very proud of the TD/18A team for the incredible work and progress they are making'

Gelsinger fires back at recent stories about 18A's poor yields, schools social media commenters on defect densities and yields : Read more

 

[why_wolf]

"While this density is slightly higher than TSMC's benchmarks at comparable stages of development,"

Why is the word "benchmarks" just linked to every Tom's article with the word benchmark in it? Completely pointless.

 

[jkflipflop98]

Yeah that "less than 10% yields" crap that's making the rounds is just that. . . crap. That's not even close to being accurate. Sad that so many news outlets have ran with this garbage.

 

[Pierce2623]

Yeah that "less than 10% yields" crap that's making the rounds is just that. . . crap. That's not even close to being accurate. Sad that so many news outlets have ran with this garbage.

As we already knew, yields depend on die size more than defect density. Like they mentioned, you’ll only get 9 perfect 5090 dies out of a wafer. Thing is Nvidia disables about 15% of the die for a 4099/5090 so most of “not-perfect” dies will still be good for 4080 super at least. Now they’re even putting 40% deactivated ad102 in 4070ti supers. So obviously those chips weren’t good enough for more expensive GPUs.

 

[waltc3]

I was under the impression that Gelsinger had resigned. Wrong impression?...

 

[phead128]

the interim Co-CEO just two days ago admitted 18A has production snags which will delay production from H1 2025 to H2 2025, presumably this is tape-out (not HVM).

So this is consistent with rumors of poor yields of 18A.

Anyways, Pat wouldn't have been fired is 18A was amazing, that's plainly obvious to anyone with half a brain.

 

[thestryker]

the interim Co-CEO just two days ago admitted 18A has production snags which will delay production from H1 2025 to H2 2025, presumably this is tape-out (not HVM).

Care to cite a source for this?

 

[watzupken]

I think anyone can try and defend, but if there is no significant takers on Intel’s 18A, it’s a pointless debate. I may not know my stuff here, but potential Intel clients will assess it themselves to see if the foundry is fit for their requirement/ expectations.

 

[cirdecus]

Isn't this copy and pasted from Yahoo?

 

[phead128]

Care to cite a source for this?

Yes, it was interim co-CEO talking during an investor presentation about 4 days ago.

https://www.investing.com/news/stoc...st-disappointing-investor-update-93CH-3757197

On Thursday, Intel Corporation (NASDAQ:INTC) experienced a decline in share value, dropping over 3% following a lackluster investor presentation held the previous day. The presentation, led by the Chief Financial Officer who has also been named interim co-CEO, failed to instill confidence among investors, according to analysts from Lynx Equity Strategies.

The analysts highlighted that the CFO's affirmation of standing by the previous earnings guidance was not persuasive, casting doubts on whether the company's quarter is on track to meet expectations. Furthermore, the anticipated progress on Intel's 18A node technology seems to have hit a snag, with the production ramp now postponed from the first half of 2025 to the second half.

The lack of clarity regarding external customers for the 18A node, coupled with the absence of concrete plans for the utilization of funds from the CHIPS Act, added to investor concerns. Additionally, the analysts pointed out vague statements about improving margins for Intel Foundry Services (IFS) next year, and a downward revision of product margin outlook, particularly for Lunar Lake, due to a price increase at competitor TSMC.

This is likely referring to 18A tape-out, which Pat said was H1 2025, which is now delayed to H2 2025. This is not likely talking about HVM, which is likely delayed to 2026. It also echos what @watzupken said, there is no major secured customers for 18A nodes, atleast not enough to get an ROI and justify the billions in fab spending. While I am not saying the yields are exactly less than 10%, I can clearly see the interim co-CEO saying 18A is having production delays and issues.

 

[Pierce2623]

Yes, it was interim co-CEO talking during an investor presentation about 4 days ago.

https://www.investing.com/news/stoc...st-disappointing-investor-update-93CH-3757197



This is likely referring to 18A tape-out, which Pat said was H1 2025, which is now delayed to H2 2025. This is not likely talking about HVM, which is likely delayed to 2026. It also echos what @watzupken said, there is no major secured customers for 18A nodes, atleast not enough to get an ROI and justify the billions in fab spending. While I am not saying the yields are exactly less than 10%, I can clearly see the interim co-CEO saying 18A is having production delays and issues.

They haven’t even taped out a chip on 18a yet? People on here were claiming Arrow Lake would use 18a. I wasn’t aware they haven’t even reached a single tape out.

 

[thestryker]

Yes, it was interim co-CEO talking during an investor presentation about 4 days ago.

https://www.investing.com/news/stoc...st-disappointing-investor-update-93CH-3757197



This is likely referring to 18A tape-out, which Pat said was H1 2025, which is now delayed to H2 2025. This is not likely talking about HVM, which is likely delayed to 2026. It also echos what @watzupken said, there is no major secured customers for 18A nodes, atleast not enough to get an ROI and justify the billions in fab spending. While I am not saying the yields are exactly less than 10%, I can clearly see the interim co-CEO saying 18A is having production delays and issues.

So an article generated by AI without an actual source cited beyond "an investor call" analysis by a finance company is your source? Congratulations on being that person. You're putting a lot of weight on something that doesn't stand up and making great leaping assumptions based on it. Also 18A has always been a 2H25 node so there's that.

Now if that article was talking about the UBS Global Technology and AI Conference call (which is the only logical conclusion) none of that is accurate. The co-CEO said nothing about delaying 18A in any fashion or even the timeline for it. The only thing related to timeline was said by Naga Chandrasekaran and here's a quote from the transcript:

Now it is about going through the remaining yield challenges, defect density challenges, continuing to improve it, improving process margin and getting it ramped. Will there be challenges? There will be, but I think we are progressing. And next year, as I look at it, primarily the first half will be getting the node into engineering samples into our customers' hands and getting the feedback and starting a ramp in Oregon. And the second half of 2025, our milestone is certifying the node, getting it ramped in Arizona and getting the product on the shelves so that customers can buy it. So that's the milestones and we are working towards meeting all those milestones over the next year. It's very critical for us.

This is the closest David Zinsner got to talking timelines, but this is in relation to foundry margins and what is driving them rather than product dates:

Now, as you point out, the next year, then we start seeing some real improvement because the next product, major product on the product side is Panther Lake. Panther Lake is an 18A -- or has 18A component in it. So we start to see wafers come back.

 

[jkflipflop98]

Like they mentioned, you’ll only get 9 perfect 5090 dies out of a wafer. Thing is Nvidia disables about 15% of the die for a 4099/5090 so most of “not-perfect” dies will still be good for 4080 super at least. Now they’re even putting 40% deactivated ad102 in 4070ti supers. So obviously those chips weren’t good enough for more expensive GPUs.

You obviously have no idea what you're talking about. Even at max reticle size, you'll get far more than 9 die per wafer. You should read more and talk less.

 

[phead128]

So an article generated by AI without an actual source cited beyond "an investor call" analysis by a finance company is your source? Congratulations on being that person. You're putting a lot of weight on something that doesn't stand up and making great leaping assumptions based on it. Also 18A has always been a 2H25 node so there's that.

Now if that article was talking about the UBS Global Technology and AI Conference call (which is the only logical conclusion) none of that is accurate. The co-CEO said nothing about delaying 18A in any fashion or even the timeline for it. The only thing related to timeline was said by Naga Chandrasekaran and here's a quote from the transcript:

This is the closest David Zinsner got to talking timelines, but this is in relation to foundry margins and what is driving them rather than product dates:

Genuine question, if 18A is so healthy, why is Pat unceremoniously fired if 18A was 2-3 quarters away from HVM (based on 0.4 defect rate)? My understanding is that tape-out has always been H1 2025, and HVM in H2 2025. A partial reason (not the main reason) Pat was fired was because of persistent yield issues with 18A resulting in tape-out delay into H2 2025 and HVM into 2026.

Btw, the investor conference was a 3 day event (Dec 2-5), so how does a 30 minute webcast on UBS website prove anything? It's possible co-CEO discussed the shifting timelines during the 3 day investor event, not during the 30 minute webcast. https://www.intc.com/news-events/ir-calendar/detail/20241204-ubs-global-technology-conference

 

[thestryker]

You obviously have no idea what you're talking about. Even at max reticle size, you'll get far more than 9 die per wafer. You should read more and talk less.

Context seems to have gone over your head as they're talking defect free die not total number or even usable.

 

[thestryker]

Genuine question, if 18A is so healthy, why was Pat unceremoniously fired if 18A was 2-3 quarters away from HVM (based on 0.4 defect rate)?

Go ask the board? There's no sign that their move was even remotely competent for the health of the company.

My understanding is that tape-out has always been H1 2025, and HVM in H2 2025. A partial reason (not the main reason) Pat was fired was because of persistent yield issues with 18A resulting in tape-out delay into H2 2025 and HVM into 2026.

Did you read the quote from Naga Chandrasekaran?

If you did you sure didn't understand it.

 

[phead128]

Go ask the board? There's no sign that their move was even remotely competent for the health of the company.

I think Pat was fired because 18A has poor yield and few major secured customers and Board was fed up with his delays. If 18A timeline is truly on-track, there is no reason to fire him.

Did you read the quote from Naga Chandrasekaran?

If you did you sure didn't understand it.

Well, surely I can be forgiven by not trusting Intel's self-professed timelines, which has consistently been delayed since 14nm and 10nm over a decade.

 

[thestryker]

I think Pat was fired because 18A has poor yield and few major secured customers and Board was fed up with his delays. If 18A timeline is truly on-track, there is no reason to fire him.

If you think there's no reason to fire him you don't understand modern wall street. Ever since Intel separated out IFS on the earnings reports the "investors" (as I refer to the short term gamblers which dominate public trading) have been calling for his head. These are the types who want money now and don't care about the health of the business. That coupled with the low margins on LNL and ARL along with the huge losses last quarter (even though most of that was one time) made those voices very loud.

I posted this in another thread regarding the board and it should tell you everything you need to know about the competence of their decisions (there's plenty before the paywall): https://www.fabricatedknowledge.com/p/the-death-of-intel-when-boards-fail

Well, surely I can be forgiven by not trusting Intel's self-professed timelines, which has consistently been delayed since 14nm and 10nm over a decade.

You can choose to not trust anything. You also need to understand there's no evidence behind anything you're saying so you should state clearly it's an opinion. Certainly don't base your opinions on something written by AI with no clear source, because at that point you're just trying to reinforce what you already believe.

 

[dalek1234]

18A yields have in fact been only 10%, but they have improved recently due to Pat and rest of Intel praying for 24 hours.