Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/threads/intel-reports-fourth-quarter-and-full-year-2024-financial-results.21983/page-2
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021770
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

Intel Reports Fourth-Quarter and Full-Year 2024 Financial Results

From Ian Cutress, Ryan Smith, More Than Moore substack:

18A Yield

"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."

"The company is noting that they're installing new equipment to further the ramp-up for high volume production, with Intel Fab 52 (Arizona) installing additional process tools to expand 18A manufacturing. Overall, Intel is expecting to kick off volume production on Intel 18A in the second half of this year, which will support the launch of Panther Lake for clients and Clearwater Forest for servers."
 
Last edited:
From Ian Cutress, Ryan Smith, More Than Moore substack:

18A Yield

"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."

"The company is noting that they're installing new equipment to further the ramp-up for high volume production, with Intel Fab 52 (Arizona) installing additional process tools to expand 18A manufacturing. Overall, Intel is expecting to kick off volume production on Intel 18A in the second half of this year, which will support the launch of Panther Lake for clients and Clearwater Forest for servers."
Just wondering: 18A in 2025 will be "Production Worthy", "Production Ready", "Production Ramp" or "High Volume Manufacturing".
 
Just wondering: 18A in 2025 will be "Production Worthy", "Production Ready", "Production Ramp" or "High Volume Manufacturing".
With Intel, the situation is somewhat more concrete: We know the fabs (AZ and Ireland), we know the products, they say they are installing the tools to meet capacity. When the ship dates are announced we'll know that. And then when the product ships to reviewers, we'll truly be in high volume manufacturing.
 
Just wondering: 18A in 2025 will be "Production Worthy", "Production Ready", "Production Ramp" or "High Volume Manufacturing".
I was wondering that as well.

Another detail I am wondering about is the size of the 18A tile(s). I was expecting that the compute tile (with memory controller) and cache tiles would be separated. Does anyone know if this is the case and what tiles would be on what process?
 
From Ian Cutress, Ryan Smith, More Than Moore substack:

18A Yield

"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."

"The company is noting that they're installing new equipment to further the ramp-up for high volume production, with Intel Fab 52 (Arizona) installing additional process tools to expand 18A manufacturing. Overall, Intel is expecting to kick off volume production on Intel 18A in the second half of this year, which will support the launch of Panther Lake for clients and Clearwater Forest for servers."

"yield curves are following standard industry trends"

Is there really such thing called "standard industry trends"?

I guess this thing can be very different from Samsung, Intel, Globalfoundries to TSMC.
 
I was wondering that as well.

Another detail I am wondering about is the size of the 18A tile(s). I was expecting that the compute tile (with memory controller) and cache tiles would be separated. Does anyone know if this is the case and what tiles would be on what process?
tile sizes: https://www.tweaktown.com/news/9936...16-cores-xe3-celestial-gpu-5-tiles/index.html
expected processes: https://overclock3d.net/news/cpu_ma...leak-hints-at-huge-celestial-xe3-gpu-upgrade/
 
"yield curves are following standard industry trends"

Is there really such thing called "standard industry trends"?

I guess this thing can be very different from Samsung, Intel, Globalfoundries to TSMC.
Intel 18A is in between Samsung 2nm on the very low end and TSMC N2 which is on the high end of the yield curve. Intel will claim 18A will ship before TSMC N2 but N2 yield is better. TSMC of course has to ship massive SoCs and GPUs in very high quantities before they make the production claim. While Intel just has to ship one of their own chip designs with chiplets from Intel and TSMC N3.
 
The issue is the company, not the CEO. Intel cannot do DCAI GPU processors... They have spend more time and money on them than nvidia. yet another "savior product" cancelled. Intel cannot do foundry (we hope to break even end of 2027) after losing 10s of billions.

Sorry Dan, Intel cannot pivot. Intel spends money wildly, moves slowly and still doesn't deliver. Remaining employees I met last week still talk about "we need more people and more money and more time" ... a new CEO will not solve the problems the company has. Jensen (my fave CEO) would be there 3 months, come up with great ideas, and the employees would say "he doesnt get how we do it here".

Focus on CPU products (Still a great leader), eliminate foundry (sell it to government contractor), make $5B per year in earnings and have a nice 10 year cruise.
 
I was at Bitmain a few years ago, and at that time, Intel released two generations of Bitcoin chips. We were very worried because Bitcoin chips use dynamic logic and domino circuits to gain an advantage in power efficiency. However, the Intel Bitcoin chips were a total failure. It seemed that the chips never leveraged the technology used in their CPU designs. That’s when we realized that Intel had significant problems, as they couldn't transfer their internal technology to another product.
Please explain how domino circuits are power efficient - I always thought that they burned more power with the need for precharge etc
 
Intel 18A is in between Samsung 2nm on the very low end and TSMC N2 which is on the high end of the yield curve. Intel will claim 18A will ship before TSMC N2 but N2 yield is better. TSMC of course has to ship massive SoCs and GPUs in very high quantities before they make the production claim. While Intel just has to ship one of their own chip designs with chiplets from Intel and TSMC N3.
Comparing with Samsung is setting the bar too low lol they consider 17mm2 dies to be good

Also I didn't get this can someone explain it me felt like it went over my head
Aaron Rakers, Wells Fargo: During 1% in 2023 EUV wafer mix, that's 5% in 2024. How do you define success what the EoY 25 in wafer mix, and the delta cost structure of EUV?
A: The blended ASP cost goes up 3x 18A vs the cost. Dramatic change vs pre-EUV wafers. Not sure exit year with % EUV, but will go up. Intel 3/4 has MTL, 18A on Panther. See a meaningful jump on EUV wafers as we exit 2025.
 
Last edited:
The issue is the company, not the CEO. Intel cannot do DCAI GPU processors... They have spend more time and money on them than nvidia. yet another "savior product" cancelled. Intel cannot do foundry (we hope to break even end of 2027) after losing 10s of billions.

Sorry Dan, Intel cannot pivot. Intel spends money wildly, moves slowly and still doesn't deliver. Remaining employees I met last week still talk about "we need more people and more money and more time" ... a new CEO will not solve the problems the company has. Jensen (my fave CEO) would be there 3 months, come up with great ideas, and the employees would say "he doesnt get how we do it here".

Focus on CPU products (Still a great leader), eliminate foundry (sell it to government contractor), make $5B per year in earnings and have a nice 10 year cruise.

The latest I hear is that Intel will in fact split manufacturing. I still think it is a mistake but what do I know. Do we really need another GlobalFoundries?

And yes there will be an Intel Foundry day this year, end of April.
 
The latest I hear is that Intel will in fact split manufacturing. I still think it is a mistake but what do I know. Do we really need another GlobalFoundries?

And yes there will be an Intel Foundry day this year, end of April.
I don't think it's new. Intel Foundry will have its own board—they mentioned that when Pat was CEO.

What DZ mentioned as the difference between the previous approach under PG and now is that capital spending is securitized and conditioned on concrete demand signals.

"Yeah. No let's see, let me go back on the capex. If you look at capex for '25, the $20 billion forecast, what's driving that lower is really a function of better utilizing assets under construction. So, our philosophy has been to invest kind of ahead of what was required, and we built up a pretty significant balance in assets under construction, I mean, to the tune of greater than $50 billion."
 
Comparing with Samsung is setting the bar too low lol they consider 17mm2 dies to be good

Also I didn't get this can someone explain it me felt like it went over my head
Aaron Rakers, Wells Fargo: During 1% in 2023 EUV wafer mix, that's 5% in 2024. How do you define success what the EoY 25 in wafer mix, and the delta cost structure of EUV?
A: The blended ASP cost goes up 3x 18A vs the cost. Dramatic change vs pre-EUV wafers. Not sure exit year with % EUV, but will go up. Intel 3/4 has MTL, 18A on Panther. See a meaningful jump on EUV wafers as we exit 2025.

Sounds like a wierd way to say margins on increases with EUV. MTL on Intel 4 is not really a big thing anymore. It has been replaced or is being replaced by Lunar lake and arrow lake. im not sure it ever outsold Raptor lake
 
I also think people are underestimating the impact of Deepseek R1 models.

If I can get it to run quite well on an Intel Arc B580, there’s no reason Intel shouldn’t be able to run the full R1 model efficiently on Intel Gaudi 3. This would be valuable for enterprise applications, as they now have local reasoning models using private data.


Jaguar Shores seems to be targeting Nvidia's core training and data center markets. To be honest, even AMD has not been able to penetrate this market effectively. Hence, I think having a second-generation product, with the first generation being Falcon Shores, makes sense.
 
Sounds like a wierd way to say margins on increases with EUV. MTL on Intel 4 is not really a big thing anymore. It has been replaced or is being replaced by Lunar lake and arrow lake. im not sure it ever outsold Raptor lake
resized.jpg

Maybe, that is expected with more utilisation of TSMC's nodes.
 
Sounds like a wierd way to say margins on increases with EUV. MTL on Intel 4 is not really a big thing anymore. It has been replaced or is being replaced by Lunar lake and arrow lake. im not sure it ever outsold Raptor lake
nothing outsold raptor lake it must have 200 Million units by now considering we have seen 3 refresh of Raptor lake(13700H,Core 100H Core 200H ) over 3 years
 
Back
Top