Industry insiders reveal that Intel's current 10 nm yields are nowhere near viable for "full production" and even with the SuperFin tech, 10 nm production may never match that of the 14 nm nodes. Plus we get more details about the performance issues with the upcoming server-grade CPUs, even more delays for the 7 nm nodes and new details about the product outsourcing plans.
https://www.notebookcheck.net/Tough-times-ahead-for-Intel-industry-insiders-spill-the-beans-on-subpar-10-nm-yields-server-CPU-plus-7-nm-delays-production-outsourcing-and-more.509532.0.html
Sic transit gloria mundi..... 8)
"7 nm is significantly more delayed than the 6-12 months first claimed. Ponte Vecchio is in-hands basically now and it was the first 7 nm product. "
you mean Intel's first 7nm product is in hand? That seems to be good news.
They also built eyeq5 chips on TSM 7nm in 2018, btw.
"Upcoming SuperFin 10 nm nodes are "unquestionably far better than base 10nm. Better in just about every way. Yields are better (50+%),"
10SF yields are 50% better than 10nm? Seems like great news. I see plenty of TGL laptops for sale. Intel announced designs for 2020 had increased from 50 to 100, so the demand could explain the shortage.
Also, Dr Su said in a recent interview that they couldn't service the peak in educational demand, while Intel stated they were no longer capacity limited. I guess the educational customers don't mind being on 14nm and AMD can't get more capacity at TSM.
"Even with SuperFin tech, the 10 nm production capacity may never match that of the 14 nm nodes."
You mean match the yields or capacity? Both are on DUV equipment. 10nm requires more DUV equipment for the multi-patterning. Never is a long time... Intel increased capacity by 25% in 2019 and again by 25% in 2020 according to recent interviews with Swan.
And what about next year's Enhanced SuperFin 10nm? and Alder-lake mobile CPUs?
Have you heard anything about it? I'm so-so curios what they will offer. I think everything will dependd on how good will be the hardware scheduler. That's why I'm struggling between Tiger-Lake H and Alder-Lake...
we shall see later
"Milan is going to likely beat Sapphire Rapids over a year earlier ..."
Milan doesn't have DDR5 support, PCIE5, CXL, Optane, AVX512, AMX tiled matrix operations, bfloat15, dlboost, right?
So, why compare it to Sapphire Rapids, which does have all those features?
If I had Intel stock, I would be selling right now. After they moved base to Israel, it seems that they've only gone downhill.
Is any credit from this reddit poster?
I worked at Intel for 17 years in Q&R. Intel persistently shortchanged Q&R. Not enough LYA (Low Yield Analysis) staff, not enough LYA equipment. So now they are dying because of low yield? They made their bed, now they are lying in it. Andy Grove is probably spinning in his grave at how stupidity has overwhelmed his company.
Quote from: JayN on December 16, 2020, 20:04:58
"Milan is going to likely beat Sapphire Rapids over a year earlier ..."
Milan doesn't have DDR5 support, PCIE5, CXL, Optane, AVX512, AMX tiled matrix operations, bfloat15, dlboost, right?
So, why compare it to Sapphire Rapids, which does have all those features?
wow it's like you actually think rattling off a bunch of buzzwords means something. do you even realize adding ddr5/pcie5 is literally just standards adherence, and everything else you listed are stupid single-customer gimmicks that no general software actually uses?
IS TSMC PLAYING WITH INTEL'S SITUATION.....
AMD'S SUDDEN DOMINANCE IS SOMEWHAT DOUBTFUL....
This article talks about 10 nm and mentions cannon lake? Wtf like y'all are so stuck on cannon lake and that debacle that you can't even analyze what is actually happening. I bet most of the people here don't even realize that the whole 14, 10, 7 nm debate is actually pointless, how can you actually talk about node process size when you don't even know what is happening? Just to make clear none of these processors are actually 14, 10 or 7 nm node if you look it's not anywhere near that also nobody considers gate, drain sizes and so on how can you talk about "node process" lithography when you don't even know what it's about? Don't even get me started on core quality when all these people out here are touting that more cores = better cpu, if that was the case AMD or ARM would have taken over a long time ago. Intel might be having a tough patch but y'all are out to lunch if you think a giant like that is just going to keel over and die.
Quote from: JayN on December 16, 2020, 20:04:58
"Milan is going to likely beat Sapphire Rapids over a year earlier ..."
Milan doesn't have DDR5 support, PCIE5, CXL, Optane, AVX512, AMX tiled matrix operations, bfloat15, dlboost, right?
So, why compare it to Sapphire Rapids, which does have all those features?
Obviously some of these "features" are not desirable for some users or even "hated" such as avx512; and AMD CPUs are not just "subset" of INTC CPUS -- they have their own features based on their customer needs.
A new CEO and some acquisitions of usable technology that assists in resolving yield issues wouldn't go amiss, along with a renewed investment in their key asset, people.
So, yes, Intel has added many features to Sapphire Rapids, and they are broadly sampling now. There have been rumors that the zen4 chip will add some of those features ... avx512, PCIE5, CXL, DDR5 ... but I haven't seen a commitment from AMD on a roadmap.
Anyway, those features are not on Milan, so my point is there is no reason to compare Milan with Sapphire Rapids' schedule.
Quote from: JayN on December 17, 2020, 15:49:18
So, yes, Intel has added many features to Sapphire Rapids, and they are broadly sampling now. There have been rumors that the zen4 chip will add some of those features ... avx512, PCIE5, CXL, DDR5 ... but I haven't seen a commitment from AMD on a roadmap.
Anyway, those features are not on Milan, so my point is there is no reason to compare Milan with Sapphire Rapids' schedule.
You have no point, it is a perfectly valid comparison. AMD is not dumb enough to waste die space with avx512, so Milan compares favorably. As for "CXL", AMD has their own high speed interconnect with a proven history of performance.
You are just spewing buzzwords with zero comprehension... here, seeking alpha, everywhere. Give it a break.
el-oh-el
It's been fun, this past year or two - batting at Intel shills and fans, who have in turn insisted that there are no 14nm capacity problems, then that Intel's 10nm has no yield issues, and then that it did but it's okay because SuperFin fixed them all, and finally that Rocket Lake will save the day and slay AMD somehow - despite clearly needing a comparatively huge die and correspondingly high power draw.
As soon as they announced the "6-12 month" delay on 7nm we all knew that would be bullshit, too, after the multiple unannounced and unspecified delays to 10nm. All but the fanboys and shills, that is.
And now it's all out there in black-and-white, straight from the horse's a**. You love to see it.
Quote from: JayN on December 16, 2020, 19:30:00
"Upcoming SuperFin 10 nm nodes are "unquestionably far better than base 10nm. Better in just about every way. Yields are better (50+%),"
10SF yields are 50% better than 10nm? Seems like great news.
In your quest to make those numbers sound good (they're not) you actually made them sound worse. The yield for OG 10nm was *less than 25%*, and that was a tiny dual-core chip with a disabled iGPU. If 10SF were only 50% better than that, you'd have yields that were under 37.5%. Contrary to your claim, that would not sound remotely like "great news". Quite the opposite.
As it is, the statement is that yields are "better than" 50%. Intel wouldn't kick themselves in the crotch by underselling this, so let's be generous and assume it's 55% (any higher and they'd say "nearly 60%"). That's dismal when you account for the fact that Intel haven't produced anything larger than 150mm^2, and explains why 8-core TGL still hasn't surfaced.
You can spin until you get dizzy if you want, but objectively speaking, this is bad.
Quote from: Some random guy on December 17, 2020, 08:04:26
This article talks about 10 nm and mentions cannon lake? Wtf like y'all are so stuck on cannon lake and that debacle that you can't even analyze what is actually happening. I bet most of the people here don't even realize that the whole 14, 10, 7 nm debate is actually pointless, how can you actually talk about node process size when you don't even know what is happening? Just to make clear none of these processors are actually 14, 10 or 7 nm node if you look it's not anywhere near that also nobody considers gate, drain sizes and so on how can you talk about "node process" lithography when you don't even know what it's about? Don't even get me started on core quality when all these people out here are touting that more cores = better cpu, if that was the case AMD or ARM would have taken over a long time ago. Intel might be having a tough patch but y'all are out to lunch if you think a giant like that is just going to keel over and die.
Amazing stuff there. Just incredible. Let's get down to brass tacks:
1) The article mentions Cannon Lake because it was Intel's first product on their 10nm process, and now we have a better idea about just how terrible the yields were.
2) It's 2020 - literally everybody knows that node names don't directly represent the feature sizes, but they *do* represent relative progress for a given foundry. They're marketing names that allow you to discuss the actual manufacturing process being used - that's all.
3) If you want to talk "core quality", M1 and Zen 3 have something to say. Maybe listen?
4) Nobody's saying Intel are going to keel over and die. It's relevant to note how many times they've screwed the pooch in the last 2-3 years, though, and that they haven't any solid plans to stop doing so.
I wouldn't give any credit for this so called "spill" first of all because of their claim that ice lake server is delayed until Q3 of 2021. It's been confirmed from multiple sources that ice lake server is going to be launched in a matter of weeks. Second point of disagreement is the current 10nm SF yield. Over a hundred different laptop models have been launched with tiger lake 10nm SF processors. I see more tiger lake laptop models than any high-end AMD laptops in stores and online. Intel is flooding the market with tiger lake laptops. There are more tiger lake laptops than all AMD laptops put together. I cannot imagine Intel has this huge imaginary yield problem and they are selling these tiger lake chips to OEMs baring huge loss from poor yields. Another fundamental thing this idiot leaker doesnt understand is yields are proportional to clock frequencies that the chip can achieve. If a high performance bin of a certain chip is clocked low in the 3ghz turbo range then you can be sure that the yields are not good. But if you see multiple bins at more than 4.3 to 4.5 ghz range then you should understand that yields of that chip are healthy. For 10nm SF I see most parts are clocked higher than 4.3ghz turbo which means the so called yield problem is a lie.
Quote from: Redbot on December 18, 2020, 22:45:18
Another fundamental thing this idiot leaker doesnt understand is yields are proportional to clock frequencies that the chip can achieve.
They are talking about functional yields, not speed binning yields. Nice try, Intel shill.
None of this matters everyone. We need 40% Female engineers and leaders. That is the only way Intel will win. This is first and only priority Intel has right now. Yield schmield. Seven schnanometer. Do you have skill and want a job? Well not if there is junk down there. Try TSMC, we are hiring in Arizona
Quote from: Some random guy on December 17, 2020, 08:04:26
This article talks about 10 nm and mentions cannon lake? Wtf like y'all are so stuck on cannon lake and that debacle that you can't even analyze what is actually happening. I bet most of the people here don't even realize that the whole 14, 10, 7 nm debate is actually pointless, how can you actually talk about node process size when you don't even know what is happening? Just to make clear none of these processors are actually 14, 10 or 7 nm node if you look it's not anywhere near that also nobody considers gate, drain sizes and so on how can you talk about "node process" lithography when you don't even know what it's about? Don't even get me started on core quality when all these people out here are touting that more cores = better cpu, if that was the case AMD or ARM would have taken over a long time ago. Intel might be having a tough patch but y'all are out to lunch if you think a giant like that is just going to keel over and die.
core quality for intel, seriously. :p :p :p :p :p :p :p intel is top down rubbish. Their cores are power hungry and even after pulling too much watts from the power supply, it cannot match Ryzen. A 10 core i9 consumes 300 watts to give worst performance than a 16 core Ryzen. Intel is overclocking their processors to match Ryzen and its all room heaters as of now. Intel is overclocking their processors and is selling as stock to get some gaming performance for kids. This intel i,dd.ii,.io,t.s, only talk about gaming. They should tell this to kids. Grown ups use CPU for many things not only gaming. But after zen 3 even that gaming claim is over. Intel fans has to understand that intel core quality is the worst. Intel is outdated crap. Only fools will buy Intel. I am happy to see this milking company is getting messed up. Apple ARM is much better and Zen 3 is far far far better.