News:

Willkommen im Notebookcheck.com Forum! Hier können sie über alle unsere Artikel und allgemein über Notebook relevante Dinge disuktieren. Viel Spass!

Main Menu

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

Started by Redaktion, December 16, 2020, 17:02:29

Previous topic - Next topic

JayN

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.

pdx

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.

Spunjji

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.

Spunjji

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.

Spunjji

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.

Redbot

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.

pdx

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.

Sob Bwan

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

ACM

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.

Quick Reply

Warning: this topic has not been posted in for at least 120 days.
Unless you're sure you want to reply, please consider starting a new topic.

Name:
Email:
Verification:
Please leave this box empty:

Shortcuts: ALT+S post or ALT+P preview