News Intel Panther Lake samples with flagship 18A node have been powered on at eight customers — Co-CEOs dispel rumors regarding poor silicon health

[Admin]

Panther Lake using Intel's 18A node has powered on at eight customers indicating the node's promising health.

Intel Panther Lake samples with flagship 18A node have been powered on at eight customers — Co-CEOs dispel rumors regarding poor silicon health : Read more

 

[Gururu]

A lot of the disinformation I am sure coming out of foreign competitors, who'd love it if the U.S. failed hard at our fabs.

 

[JRStern]

Better than nothing.
Still a possibility it will win, and 18a will at least equal anything from TSM.
But, ongoing yield and cost issues will continue to haunt everyone.

 

[TheSecondPower]

What's the usual time between OEMs getting engineering samples and laptops reaching the market? Like 6-9 months or something?

 

[thestryker]

PTL/CWF are both very important from a product and foundry standpoint. PTL should bring Intel's margins back in line with where they expect to be while also being a client showcase for 18A. CWF while not as important on the margin side (all Xeons are built on Intel nodes) it will be extremely important for the enterprise market in scale and performance. From a foundry perspective I don't think this will end up being much different size wise than PTL since all available information indicates 24 core tiles.

What's the usual time between OEMs getting engineering samples and laptops reaching the market? Like 6-9 months or something?

That sounds about right, but in this case I'd assume we're talking mid-late fall releases at the earliest since ARL laptops won't launch until next year. While I'm sure the OEMs and Intel would both like to have best foot forward they also want to be able to sell existing stock.

I do think that the ARL laptop chips will probably be the least produced of recent history due to PTL seemingly being on track and the cost of TSMC N3 manufacturing.

 

[Mama Changa]

What's the usual time between OEMs getting engineering samples and laptops reaching the market? Like 6-9 months or something?

Panther Lake is due for release in H2 2025, but probably late Q3 if things go to plan. Also, let's not forget their server cpu Clearwater Forest is also built on 18A and due next year.

 

[DaveLTX]

A lot of the disinformation I am sure coming out of foreign competitors, who'd love it if the U.S. failed hard at our fabs.

Disinformation? You mean reality?
Intel 20A was supposed to be the first customer node and they dropped it. Doesn't instill confidence that we know that all intel really can do is for themselves... And why ARL went to N3?

 

[thestryker]

Disinformation? You mean reality?

There's zero evidence supporting the claims against 18A being "reality". Most of the reporting has been dishonest at best. While I don't think there's some grand conspiracy it's very likely that at least some of it is being done by investors trying to maximize short term profits.

Intel 20A was supposed to be the first customer node and they dropped it.

This was supposed to be a one and done node which means their reasoning for dropping it is believable, especially after Intel 4. However until 18A appears nobody will know whether or not this was the real reason. That's the only reality of the situation: Intel has stated their reason, but that's the only actual information available.

And why ARL went to N3?

It was supposed to be on 20A, but the node was canceled and LNL was already on N3. In theory this makes doing ARL on N3 faster than making it work on Intel 3 and frees up EUV capacity in the process. Assuming 18A is actually hitting its timelines this is the most logical approach for the long term health of IFS.

 

[SunMaster]

A lot of the disinformation I am sure coming out of foreign competitors, who'd love it if the U.S. failed hard at our fabs.

Oh, is it a competition where Intel is synonomous with 'US"? And 'our" fabs are manufacturing using a national secret, I presume,

 

[kjfatl]

What's the usual time between OEMs getting engineering samples and laptops reaching the market? Like 6-9 months or something?

6-9 months for a new part of this complexity on a new process seems like an aggressive schedule. If everything goes right it is reasonable, but I would expect evaluation phases and 2 or 3 loops through the fab to fix bugs and make changes to optimize process yields. This puts it at 9 to 15 months. First parts came off of the fab over a month ago.

 

[phead128]

A lot of the disinformation I am sure coming out of foreign competitors, who'd love it if the U.S. failed hard at our fabs.

Engineering samples powering on doesn't imply anything about yield at HVM scale though.

Pat Gelsinger famously handed an 18A engineering sample to Lenovo CEO, then he got fired. So what does that say about 18A health with just passing out initial samples to customers?

Pat shows engineering sample of 18A to Lenovo just four weeks before he got fired.

TLDR: engineering samples does not imply healthy HVM yields at scale.

 

[jp7189]

6-9 months for a new part of this complexity on a new process seems like an aggressive schedule. If everything goes right it is reasonable, but I would expect evaluation phases and 2 or 3 loops through the fab to fix bugs and make changes to optimize process yields. This puts it at 9 to 15 months. First parts came off of the fab over a month ago.

Those fabs and all the staff are focused exclusively on this one product without distractions. I'm rooting for them, though my expectations are tempered by recent poor performance.

 

[Gururu]

Oh, is it a competition where Intel is synonomous with 'US"? And 'our" fabs are manufacturing using a national secret, I presume,

Intel is but another domino in what is so very much a strategic game of world domination hinged on technological advancement. You bet your butter foreign interests are in full gear to see Intel dismantled.

 

[phead128]

Intel is but another domino in what is so very much a strategic game of world domination hinged on technological advancement. You bet your butter foreign interests are in full gear to see Intel dismantled.

Irony after US tried to kill Huawei and SMIC, it's crying that it gets a taste of its own medicine.

 

[dCasualGamer]

Disinformation? You mean reality?
Intel 20A was supposed to be the first customer node and they dropped it. Doesn't instill confidence that we know that all intel really can do is for themselves... And why ARL went to N3?

Reality is that there are timelines and hard data like defect per area that is being reported out. Taking the fact and put some personal judgement to it can be disinformation. Taking an example, Intel has published a defect density of 0.4/cm^2 and that is inline with industry norm. That is a fact, with no emotion and personal bias/judgement. Taking that fact and then saying things like "... TSMC has better number, so Intel is so dead behind..." or "... I don't trust what they said, so that number must be fake..." are examples of disinformation.

Why ARL goes to N3 ? It takes 3-4 years to design and produce a chip. It's not like Intel can just decide to switch to the 18A process. The 18A process and ARL are not even in the same timeline.
You are implying that Intel had a choice between 18A/20A and TSMC and Intel decided to go with TSMC, hence that implies that something sneaky is happening. That's disinformation.

 

[phead128]

Disinformation? You mean reality?
Intel 20A was supposed to be the first customer node and they dropped it. Doesn't instill confidence that we know that all intel really can do is for themselves... And why ARL went to N3?

You know what really instills confidence? The article says 30% of 18A will be outsourced to TSMC, only 70% of 18A will be in-house manufacturing (source). I predict that similar to 20A, they will outsource more and more of 18A to TSMC as Intel IFS can't pick up the slack. It's almost predictable at this point with Pat's ouster and shifting to IFS spinning-out model.

 

[thestryker]

You know what really instills confidence? The article says 30% of 18A will be outsourced to TSMC, only 70% of 18A will be in-house manufacturing (source). I predict that similar to 20A, they will outsource more and more of 18A to TSMC as Intel IFS can't pick up the slack. It's almost predictable at this point with Pat's ouster and shifting to IFS spinning-out model.

Yeah you don't seem to be understanding what that quote is referring to at all. It's referring to the titles being used in the PTL CPU. Chances are that the Graphics tile is still going to be TSMC and depending on the overall CPU design more tiles may as well. For all of their existing client CPUs with tiles only the Compute tile and/or Base tile are Intel (the latter doesn't really count).

 

[phead128]

Yeah you don't seem to be understanding what that quote is referring to at all. It's referring to the titles being used in the PTL CPU. Chances are that the Graphics tile is still going to be TSMC and depending on the overall CPU design more tiles may as well. For all of their existing client CPUs with tiles only the Compute tile and/or Base tile are Intel (the latter doesn't really count).

I get what you're saying about the tiles, but this goes deeper than just Graphics vs Compute tiles. The real issue is Intel's growing TSMC dependency, even for their supposed crown jewel nodes like 20A and 18A.

Let's be real - if Intel can only manage 70% in-house production for 18A (according to the article), that's a red flag for IFS's capabilities. Their track record with 20A hasn't exactly inspired confidence either.

With all the management shake-ups lately and these production issues, I wouldn't be surprised if we see Intel leaning even more on TSMC down the road. They can talk about sovereignty and in-house production all they want, but the numbers tell a different story.

Just my 2 cents, but IFS needs to prove they can actually deliver before we buy the whole 'foundry renaissance' narrative.

 

[dCasualGamer]

Let's be real - if Intel can only manage 70% in-house production for 18A (according to the article), that's a red flag for IFS's capabilities. Their track record with 20A hasn't exactly inspired confidence either.

With the volume that Intel ships each year, I would call a 70% in-house production a WIN.

 

[phead128]

With the volume that Intel ships each year, I would call a 70% in-house production a WIN.

Intel Products ain't going to save Intel IFS Foundry, they are explicitly looking at external customers. If Intel Products can support the ROI on 18A, then why go for a Foundry model? To achieve sustainability and growth, IFS must secure a diverse customer base beyond internal Intel demands - that's Pat Geisinger's IDM 2.0 strategy in a nut shell.

 

[thestryker]

Let's be real - if Intel can only manage 70% in-house production for 18A (according to the article), that's a red flag for IFS's capabilities. Their track record with 20A hasn't exactly inspired confidence either.

All of Intel's Xe core graphics releases have been using TSMC. The goal of the graphics hardware team is also yearly refreshes which means Xe3 is done and they're working on Xe4. Keeping timelines here could easily limit the ability to shift nodes, or make them cement the node early on.

Intel also has limited EUV capacity to an extent since High-NA was delayed and they're having to spin up fabs. IFS also doesn't really have a cost effective node between Intel 16 and Intel 3. They may end up manufacturing some parts on Intel 7, but everything I've seen indicates it's more expensive than Intel 3.

It's really not the red flag you seem to think it is based on the information available now. I'm not saying there's nothing to be worried about just that the available evidence doesn't warrant concern.

 

[dCasualGamer]

Intel Products ain't going to save Intel IFS Foundry, they are explicitly looking at external customers. If Intel Products can support the ROI on 18A, then why go for a Foundry model? To achieve sustainability and growth, IFS must secure a diverse customer base beyond internal Intel demands - that's Pat Geisinger's IDM 2.0 strategy in a nut shell.

New process node always need its Customer-0, the first to use the process and pay for all the inefficiency and learning. Apple is TSMC Customer-0. Since Intel 18A is the industry newcomer, then Intel Product will be the Customer-0 - to clean up thew process as well as serving as a testament to other companies of its viability.

As it has been mentioned earlier, a chip design takes 3-4 years. That means
1) Intel is saying Pantherlake is on 18A. That is a decision few years back. Whatever Intel product on TSMC today is decision made >5 years ago.
2) Companies wouldn't want to bet their future 3-4 years down the road on an unproven process. So Intel product as 18A Customer-0 is important for other companies to have the confidence to bet their products, 3-4 years down the road are good

 

[phead128]

New process node always need its Customer-0, the first to use the process and pay for all the inefficiency and learning. Apple is TSMC Customer-0.

Apple as Customer-0 guarantees money to scale fabs, but does not guarantee good yields. For instance Apple dual-sourced 14nm A6 chip from Samsung and TSMC, yet Samsung flopped so badly, TSMC won all orders for A10 chip because of its superior yields over Samsung. In other words, deep pockets does not preclude poor execution or guarantee good yields.

Since Intel 18A is the industry newcomer, then Intel Product will be the Customer-0 - to clean up thew process as well as serving as a testament to other companies of its viability.

Intel tried to be a foundry in 2012, which failed in 2018 because it only had Intel Products as it's main customer. This is now Intel's second attempt at becoming a foundry, relying solely on Intel Products will have predictable results like the first attempt failure.

As it has been mentioned earlier, a chip design takes 3-4 years. That means
1) Intel is saying Pantherlake is on 18A. That is a decision few years back. Whatever Intel product on TSMC today is decision made >5 years ago.
2) Companies wouldn't want to bet their future 3-4 years down the road on an unproven process. So Intel product as 18A Customer-0 is important for other companies to have the confidence to bet their products, 3-4 years down the road are good

We know Intel Products is a customer of Intel Foundry, that's an open secret. The key to Intel IFS foundry's survival is if it can secure major customers, because $100 billion in fab expansions is not sustainable on internal Intel demand alone. It defeats the point of a Foundry model by just serving Intel Products, just stay as an IDM if you think that's a "win".

 

[dCasualGamer]

Apple as Customer-0 guarantees money to scale fabs, but does not guarantee good yields. For instance Apple dual-sourced 14nm A6 chip from Samsung and TSMC, yet Samsung flopped so badly, TSMC won all orders for A10 chip because of its superior yields over Samsung. In other words, deep pockets does not preclude poor execution or guarantee good yields.

Definitely agree that yield or defect density is a not a guarantee. However, you need to have the Customer-0 order to have the factory running, to learn from issues and to continue improving.

We know Intel Products is a customer of Intel Foundry, that's an open secret. The key to Intel IFS foundry's survival is if it can secure major customers, because $100 billion in fab expansions is not sustainable on internal Intel demand alone. It defeats the point of a Foundry model by just serving Intel Products, just stay as an IDM if you think that's a "win".

While it is not a guarantee that other companies will follow Intel Product to make IFS a true foundry, not doing it is a guarantee of failure. No many customers will risk on an unproven process nor have the financial means to be Customer-0.

 

[ToBeGood]

Intel Products ain't going to save Intel IFS Foundry, they are explicitly looking at external customers. If Intel Products can support the ROI on 18A, then why go for a Foundry model? To achieve sustainability and growth, IFS must secure a diverse customer base beyond internal Intel demands - that's Pat Geisinger's IDM 2.0 strategy in a nut shell.

Intel Product "alone" is going to save Intel Foundry. Do you know that Apple is like 40% of all TSMC revenue in 2022.

Do you know how many things that intel is FABing / making.

80% of Laptop CPU.
50% of Server CPU.
All the chipset mentioned above
In 2025, I am estimating that Arc Battlemage should have 15% of GPU. Look at B580.

That is already at least if not larger than Apple, i.e. 40% of TSMC revenue of 2022. Of course, now Intel is fab at TSMC temporary so Apple Share is lower.
Vs Apple
5% of Laptop CPU
50% of Mobile Phone SoC

Vs AMD
10% of Laptop CPU
70% of Desktop CPU
But Laptop Vs Desktop sales is around 3: 1
The internet really wants Intel goes down but in fact Intel is all the way up and up.

Intel just received $7.6 Billion pure cash (so close to profit) from the US government for IFS.

The other funny things is that Intel Arc B580 is making a loss based on Bill of Material.

I don't know this is a good source but someone is saying this:

https://www.quora.com/How-much-is-t...If-it-is-how-much-do-you-think-it-should-cost

If nVidia is paying only 200 USD for a 600mm^2 die, how much intel going to pay for B580 272mm^2 die, with 2-4 times more money Intel pay to TSMC vs nVidia pay to TSMC, plus based on a N4/5 which is not leading node i.e. N3E/B ???? that node is already 4 years old my friend. 80 USD my friend is the amount we are thinking of plus the PCB, packing, shipping, etc. 180 USD is fair i.e. intel and its partner is making 70 USD per card, not very bad.

It is very reasonable to think that nVidia is making 40% GP per GPU card sell on Recommended Retail Price (4060) . The Cost of Material of nVidia and its broad partner should be around 180 USD. While Intel is selling the card at 250 USD.

Don't be fool, Intel does make a loss for GPU but not variable cost, Fixed Cost and the whole division as a whole Intel might still be making loss, i.e. include cost of driver development, game studio deployment, rebate of partners i.e. Acer (as Acer is a major laptop maker that is also why Acer is the once promote Arc so hard since they can get rebate on other Intel products i.e. Laptop CPU). But purely on Cost of Material, I don't think Intel is making a loss.

That is why they are saying they are ramping up production, with weekly stock delivery, a clear sign that they are making money per CPU. Don't Trust MLID.

Talking about competitiveness, I do think that the hardware alone, B580 have similar transitors count Vs 4070, then should intel worry about 5060 vs B580, nope, based on what we saw so far, B580 should at least have another 15% performance improvement in the next 6 months (on pure driver alone), although 5060 will have 50% improvement, but let see if 5060 still stuck at 8G VRAM, sorry, 1440p is not going to go well. 5060 is GDDR 7 and B580 is GDDR 6 so nVidia is not going to sell 5060 for less then 350 USD, if not Battlemage, there are still going to be tons of people buying 4060, so now those people having 1 more option and that is B580.