• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Intel 18A Process Node Clocks an Abysmal 10% Yield: Report

erek

erek

Fully [H]
2FA
Joined
Dec 19, 2005
Messages
17,664
Local references: “Intel CEO admits 'I've bet the whole company on 18A'
"I’ve bet the whole company on 18A." So, says IntelCEO Pat Gelsinger in the most deadpan, matter-of-fact manner imaginable. He's not glib. He's not joking. Everything rides on Intel's 18A process according to Gelsinger.”

—-
1733401575962.png

“The report also hints that Intel's in-house foundry nodes going off the rails could be an important event leading up to the company's Board letting go of former CEO Pat Gelsinger, as huge 2nd order effects will be felt across the company's entire product stack in development. For example, company roadmaps put the company's next-generation "Clearwater Forest" server processor, slated for 2025, as being designed for the Intel 18A node. Unless Intel Foundry can pull a miracle, an effort must be underway to redesign the chip for whichever TSMC node is considered cutting-edge in 2025.”

Source: https://www.techpowerup.com/329513/intel-18a-process-node-clocks-an-abysmal-10-yield-report
 
Marees said:
How reliable is Chosun ??

10% seems incredibly low
Click to expand...
In the past they publish a lot of pro Korea anti everyone else hit pieces.

I’d believe it to be low but not 10%, likely closer to 40%.

But Toms published this not too long ago.

https://www.tomshardware.com/tech-i...8a-is-healthy-potential-clients-are-lining-up

"I am happy to update the audience that that we are now, for this production process, we are now below 0.4 d0 defect density, this is now a healthy process," said Pat Gelsinger, chief executive of Intel, at the technology conference.


When it comes to defect density (D0), then it is commonly believed that a defect density below 0.5 defects per square centimeter is good (0.5 def/cm^2), so Intel 18A's 0.4 defects per cm^2 is a good result (at least on paper), especially considering the fact that the production technology is a few quarters away from entering mass production, at which point defect density will likely get lower.
 
Even if the yields are low, they should get better. Analyze the defects, put in more dust filtering.
 
Halon said:
Rumors are allegedly tied to Broadcom, who has an interest in a hostile takeover. Take with a grain of salt, especially given the prior update from Tom’s.
Click to expand...
You don’t trust tom’s?
 
While I certainly wouldn't put it past Intel to have another disaster of a node... I also am hesitant to believe random Interweb rumors, especially when said rumors align with the current zeitgeist of who's the "good guys" and "bad guys".

As with any other news about things that haven't happened yet I'm always wait and see. I've seen entirely too many rumors that were either far too optimistic, or far too doom-y compared to what actually ended up happening to believe shit before it is official.

My favorite from years past was the transition to DX10 cards. The rumor mill was that AMD would release the first DX10 part, because of their experience working with MS on the Xbox 360 and it would be true DX10 hardware with unified shaders. Big bad nVidia would be later, and their would support DX10 but would still be separate pixel and vertex shaders underneath and not nearly as performant. Then, like a bolt of lightning out of a clear blue sky, the 8800 drops and it is not only the first DX10 part, but it just completely blows the doors off of anything else performance wise even in DX9 titles. The rumor mill was sure it would be a "doom for nVidia" thing and it was just the opposite.

Likewise I've seen way too many things overhyped that either were not as good or just failed to materialize to get hyped about something not on the market yet.
 
Lakados said:
In the past they publish a lot of pro Korea anti everyone else hit pieces.

I’d believe it to be low but not 10%, likely closer to 40%.

But Toms published this not too long ago.

https://www.tomshardware.com/tech-i...8a-is-healthy-potential-clients-are-lining-up

"I am happy to update the audience that that we are now, for this production process, we are now below 0.4 d0 defect density, this is now a healthy process," said Pat Gelsinger, chief executive of Intel, at the technology conference.


When it comes to defect density (D0), then it is commonly believed that a defect density below 0.5 defects per square centimeter is good (0.5 def/cm^2), so Intel 18A's 0.4 defects per cm^2 is a good result (at least on paper), especially considering the fact that the production technology is a few quarters away from entering mass production, at which point defect density will likely get lower.
Click to expand...
40% is just as disastrous. You can't make profitable parts with even a 25% defect rate. A 60% defect rate is trash.
I mean 90% defective seems very high... though I wouldn't be shocked if at some point there was a run or two at Intel that was just that bad.
 
defect rate (when not said by 100mm square, but in %) depends on chip size, so without that value can get hard to judge a bit.

40% on 800m die could be really good (biggest chips broadcomm), complete disaster for 70mm ?

Cannot find the actual article that mention 10%
 
erek said:
You don’t trust tom’s?
Click to expand...
No, the opposite - Tom’s published a piece of news with initial positive indicators, this piece appears to refute that. I am more skeptical of the leak from today. Sorry for the confusion.
 
Halon said:
this piece appears to refute that
Click to expand...
Does it ?

According to this:
https://semianalysis.com/die-yield-calculator/

A 0.4 defect rate per cm2 claimed by intel would give a 9% yield for a 800mm chips, TSMC 0.1 for mature N5 would be 47%

And your failure rate can be affected by the chips you do I imagine, not just the size of it ?

Both party could be saying complete factual affair that do not contradict each other, if broadcomm wanted to use Intel for very large chips, what does 10% yield even mean if we do not know the size and nature of the chips they are trying to make.... can mean good for other things 0.4 defect by cm, could mean unimaginable bad 3 error by cm for phone soc,
 
GoldenTiger said:
This, and to have a good chip company in general competition. Not sure why anyone would root for Intel to fail.
Click to expand...
Intel has stifled innovation for 35 years. Its not hard to understand how a lot of people are smiling ear to ear seeing them flame out.

I guess I want to see them turn it around at some point. I'm not convinced that is going to happen. I'm not hoping for it though. They do have many decades of bad karma.
 
zandor said:
Why?

I'm pulling for Intel to get back in the game because of USA fabs.
Click to expand...
I am a 13900k owner. For months i thought it was something i did in setting and my computer kept crashing and crashing to the point I build a 2nd computer. I am going to send a e-mail with my info and see what they say for a rma. I hope its easy to get a rma. But yes fuck intel for the bull shit chip i paid like $700 for back in the day.
 
ChadD said:
Intel has stifled innovation for 35 years.
Click to expand...
Then buy from AMD and other vendors.

ChadD said:
Its not hard to understand how a lot of people are smiling ear to ear seeing them flame out.
Click to expand...
If they flame out, then there is no competition for AMD. And then AMD will grow fat, dumb and happy. Which would be bad for all of us.,
ChadD said:
I guess I want to see them turn it around at some point. I'm not convinced that is going to happen. I'm not hoping for it though. They do have many decades of bad karma.
Click to expand...
Yes, decades of being stupid in different ways. And so now they are paying. I just hope that Intel purges its ranks of the people in power who were responsible for being "stupid.:" But Like ChadD I'm not optimistic. It would not entirely surprise me if some private equity company makes a bid for Intel. if they have the balls to do it, they might offer shareholders less than current market price for their stock, so the numbers work out for them. Then they will part out the company.
GoldenTiger said:
Oh, I see... We're doing the revisionist history thing again.
Click to expand...
What is so revisionist?
 
Epyon said:
I am a 13900k owner. For months i thought it was something i did in setting and my computer kept crashing and crashing to the point I build a 2nd computer. I am going to send a e-mail with my info and see what they say for a rma. I hope its easy to get a rma. But yes fuck intel for the bull shit chip i paid like $700 for back in the day.
Click to expand...
Why didn't you rma what you thought was a faulty chip, rather than build an entirely new computer? /boggle
 
Comment by ex-AT CPU reviewer

When you don't understand the difference between defect rate and yield. Pat said in August the defect rate was below 0.40 - at reticle sized chips, that's an 8% yield - but at smartphone sized chips, that's ~65%. There's a reason new nodes have pipecleaners that are small. Also, still many months out from actual HVM.​
​
If having sub-10% yield is the take-away, that's the story you could have had in August (it was wrong to charactize the process like that back then, too). Why this decides to flair up now is because sounds like an idiot investor planting rumors. AGAIN.​

https://x.com/IanCutress/status/1864819463184409074
 
Cutress is right... I'm sure 18A is several months away from being used for production. There's no reason to panic now — I'd only really be worried if the results were this low with customer output just a month or two away.
 
philb2 said:
If they flame out, then there is no competition for AMD. And then AMD will grow fat, dumb and happy. Which would be bad for all of us.,
Click to expand...
Completely untrue - this isn't 2005. ARM has established beach heads in every traditionally-x86 market. That's all the competition AMD needs to keep on the accelerator.

Let it burn.
 
Very foolish and strange behavior.
He should have understood that his job was to both innovate upon technology, and make money.
And he will end up doing neither.
 
Twisted Kidney said:
10% is not at all bad for a developing node.
Click to expand...
Could be incredibly terrible, could be not that bad.... how does one know without the size of the chips (and the nature of the chips itself, if those network chips are like gpu and you can have only 88% of redundant subsection working or something different that need to error on all of it) ? It is a very low information number by itself, never used in a vaccum like that.

philb2 said:
If they flame out, then there is no competition for AMD. And then AMD will grow fat, dumb and happy. Which would be bad for all of us.,
Click to expand...
We can imagine the state forcing AMD to make a deal for the x86_64 licensing with the people that bought Intel (or their patents).
 
Last edited:
Aurelius said:
Cutress is right... I'm sure 18A is several months away from being used for production. There's no reason to panic now — I'd only really be worried if the results were this low with customer output just a month or two away.
Click to expand...
I believe I read that 18A is to go into mass production at this time next year? So there is still plenty of time to improve the process. 18A is the is the 'P' in Intel's Process-Architecture-Optimization cycle, so it won't be a high-volume product, anyway. Intel 3 will be the high-volume product next year. 14A in 2027 will be the one to watch for in the desktop market.
Centauri said:
Completely untrue - this isn't 2005. ARM has established beach heads in every traditionally-x86 market. That's all the competition AMD needs to keep on the accelerator.

Let it burn.
Click to expand...
Maybe in the server market and growing in the laptop market. As a gamer in the desktop market I will never be in the market for an arm CPU.
 
ChadD said:
40% is just as disastrous. You can't make profitable parts with even a 25% defect rate. A 60% defect rate is trash.
I mean 90% defective seems very high... though I wouldn't be shocked if at some point there was a run or two at Intel that was just that bad.
Click to expand...
I said 40%, but when measuring defects per square mm, the actual rate is very dependent on the size of the chip
So "production ready" is considered around 0.5, Intel is just below 0.4, TSMC launched the 5N process very successfully at 0.33, and it only improved from there as they tuned it in.

Basically if Intel was still doing monolithic designs 0.4 would be pretty OK, but it would result in SKU hell, which is what we sort of expect from Intel, but if you ever wondered why they have so many SKU's for their chips its because they waste very little silicon.
Intel figures out every permutation of where an error could be in a chip and still have it function, then creates a SKU for it and during the binning process they simply kill off the offending sections or clock them down accordingly to make them work then bin it with those specs.

Chiplets or Tiles or whatever you want to call them means smaller chips which means even at 0.4 the chances of multiple defects in a single unit is going to be rare, which ultimately means fewer bins with fewer SKUs.
If you combine that with the much smaller max reticle size that the new ASML hardware Intel is running works at the process itself requires chips to be much smaller, roughly half the max size of the older model of ASML hardware TSMC uses for their 3N process.
 
ChadD said:
Intel has stifled innovation for 35 years.
Click to expand...
GoldenTiger said:
Oh, I see... We're doing the revisionist history thing again.
Click to expand...
philb2 said:
Then buy from AMD and other vendors.
Click to expand...
It is absolutely not revisionist to say that Intel has been preventing innovation from reaching users for decades. One major method was by making competitor's products harder to buy.
From EU antitrust case against Intel (these facts were not disputed, unlike the fine):

First, Intel gave wholly or partially hidden rebates to computer manufacturers on condition that they bought all, or almost all, their x86 CPUs from Intel. Intel also made direct payments to a major retailer on condition it stock only computers with Intel x86 CPUs. Such rebates and payments effectively prevented customers - and ultimately consumers - from choosing alternative products. Second, Intel made direct payments to computer manufacturers to halt or delay the launch of specific products containing competitors' x86 CPUs and to limit the sales channels available to these products. (emphasis mine)​

And such practices didn't stop then. When AMD launched Zen into the server market, Intel specifically went to undercut AMD by authorizing OEMs to offer larger rebates for customers who considered AMD products:

A key trigger seems to be an organization’s willingness to adopt AMD EPYC. Our advice, if you are buying even as few as 50-100 servers, is to get an AMD EPYC system quote from your reseller. Doing so seems to be the trigger for Intel’s discount approvals.​

It is clear that Intel does not and cannot compete on merit. The inability to drive CPU performance forward was also what drove Apple away, once they realized that they could to much better with their own designs on TSMC manufacturing:
perf-trajectory_575px.png

Source: https://www.anandtech.com/show/16226/apple-silicon-m1-a14-deep-dive/4
 
Lakados said:
I said 40%, but when measuring defects per square mm, the actual rate is very dependent on the size of the chip
So "production ready" is considered around 0.5, Intel is just below 0.4, TSMC launched the 5N process very successfully at 0.33, and it only improved from there as they tuned it in.

Basically if Intel was still doing monolithic designs 0.4 would be pretty OK, but it would result in SKU hell, which is what we sort of expect from Intel, but if you ever wondered why they have so many SKU's for their chips its because they waste very little silicon.
Intel figures out every permutation of where an error could be in a chip and still have it function, then creates a SKU for it and during the binning process they simply kill off the offending sections or clock them down accordingly to make them work then bin it with those specs.

Chiplets or Tiles or whatever you want to call them means smaller chips which means even at 0.4 the chances of multiple defects in a single unit is going to be rare, which ultimately means fewer bins with fewer SKUs.
If you combine that with the much smaller max reticle size that the new ASML hardware Intel is running works at the process itself requires chips to be much smaller, roughly half the max size of the older model of ASML hardware TSMC uses for their 3N process.
Click to expand...
I think the issue for Intel... is they are selling fab space. I'm not in the field, it seems to me TSMC spends a lot of time and money helping their customers out with design and layout to best improve yields. I know of course Intel is doing the same, they just have a lot less experience with it. Intels fab people frankly are used to having things go wrong and just pushing through millions in silicon anyway... cause its all Intel. You can't do that with customers. You either cost them money or your profits. I could be wrong, I just get that feeling so far everyone that has worked with Intel hasn't seemed to be all that happy.
 
ChadD said:
I think the issue for Intel... is they are selling fab space. I'm not in the field, it seems to me TSMC spends a lot of time and money helping their customers out with design and layout to best improve yields. I know of course Intel is doing the same, they just have a lot less experience with it. Intels fab people frankly are used to having things go wrong and just pushing through millions in silicon anyway... cause its all Intel. You can't do that with customers. You either cost them money or your profits. I could be wrong, I just get that feeling so far everyone that has worked with Intel hasn't seemed to be all that happy.
Click to expand...
Intel's current design software is reportedly trash, and it was 6 months late.
 
  • Like
Reactions: ChadD
like this
If people don't want monopolistic prices to keep on getting higher, then what you don't want is Intel to fail, as well as TMSC being the only ones making high-end chips. All 3, AMD, Intel, and NVIDIA have done shady stuff, but if you like gaming computers, they're the only game in town... for now.
 
El_Capitan said:
If people don't want monopolistic prices to keep on getting higher, then what you don't want is Intel to fail, as well as TMSC being the only ones making high-end chips. All 3, AMD, Intel, and NVIDIA have done shady stuff, but if you like gaming computers, they're the only game in town... for now.
Click to expand...
I've never understood why people think some companies are good and some are evil and if only the evil ones went away, everything would be great. No, all companies are greedy. It is kinda their mandate. What that means is if you have a "last man standing" situation they are going to take advantage of it, even if they were supposedly one of the "good" ones. Doesn't matter if it is AMD, TSMC, or whoever you think is the "underdog" if they become the only game in town prices will go up, innovation will go down, and it'll suck more for consumers.

So really, no people, you DON'T want Intel gone. If that happens then AMD is it for x86 CPUs and while TSMC isn't completely it for high end fabs, they will be extremely dominant. That is not a situation that will make things better for you. You think AMD is going to work so hard on making better Zen CPUs if they are all you can buy? You think prices will get any better is TSMC is the only fab around?
 
Armenius said:
Maybe in the server market and growing in the laptop market. As a gamer in the desktop market I will never be in the market for an arm CPU.
Click to expand...
Newsflash: The PC is no longer commanding sales numbers that justify expensive development of CPUs just for it.

AMD uses since 2017 (Zen) repurposed server silicon for their higher performing processors on the desktop. For notebooks and tablet PCs, we are seeing repeated attempts by Microsoft since Windows RT (2013) to prepare for the era of repurposed mobile processors. For Apple macOS (and also Google Chrome OS), this era is already here.

View: https://www.youtube.com/watch?v=_XNLXQ7_bs4

If/when ARM comes to dominate in servers, then you won't have a choice any more, and live with x86 emulation for your legacy games. This is also why Intel going away would not mean the end of competition in the desktop market.

El_Capitan said:
All 3, AMD, Intel, and NVIDIA have done shady stuff,
Click to expand...
NVIDIA has done really shady stuff and is up there with Intel and Microsoft as the three most evil and anti-competitive companies to ever exist in tech. But at least they innovate and relentlessly work on delivering bigger and better gear to their customers (and make them pay through the nose for it).

AMD on the other hand does not meet the definition of shady. Yes they made many stupid moves and are among the least agile in the industry. But even when AMD was in the performance lead, their product prices were always fair, and they never engaged in anti-competitive practices.
Sycraft said:
I've never understood why people think some companies are good and some are evil and if only the evil ones went away, everything would be great. No, all companies are greedy. It is kinda their mandate.
Click to expand...
There exists something called corporate culture. All companies want to make money, sure, but the way they go about it is vastly different.

Intel is evil from the core and used to getting their products into people's hands even if inferior, if necessary by paying money to lock out competitors and making back that money when customers no longer have a choice as result.

We can all be happy that this didn't work in mobile (google "Intel contra revenue"), and now this scheme also collapsed in desktop and server markets as Intel went too long without competitive product.
 
Armenius said:
Maybe in the server market and growing in the laptop market. As a gamer in the desktop market I will never be in the market for an arm CPU.
Click to expand...
I won’t be surprised if ARM has perpetual trouble in the desktop gaming space, but… never? You sure?

Remember, there have been many times where people would have laughed if you’d said AMD would clobber Intel in most cases. What seems utterly stable now can change if the conditions are right.

As it is, GPU prices are getting to the point where even some serious PC gamers are having second thoughts about buying the higher end cards.
 
You must log in or register to reply here.
Back
Top