• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

cannon Lake delayed (again)

Some of us were talking about it earlier and holiday season means November 2019,so 2H 2019 is probably more like Q4 2019 unless Intel makes some breakthrough before then.
 
It's quite shocking when you think about Intel initially planned to release their first 10nm CPU's in 2016!
The crazy thing is we are likely facing a situation were AMD will leapfrog Intel with their 7nm die shrink at a time Intel remains stuck on 14nm desktop CPU's that's sprinkled with security issues on top :eek:
 
NinjaCool said:
It's quite shocking when you think about Intel initially planned to release their first 10nm CPU's in 2016!
The crazy thing is we are likely facing a situation were AMD will leapfrog Intel with their 7nm die shrink at a time Intel remains stuck on 14nm desktop CPU's that's sprinkled with security issues on top :eek:
Click to expand...
Is not only desktop CPUs. Is the Intel servers also stuck.
And the new HEDT CPUs are 14nm also. Chillers will be the new AIOs soon.... :D
 
NinjaCool said:
It's quite shocking when you think about Intel initially planned to release their first 10nm CPU's in 2016!
The crazy thing is we are likely facing a situation were AMD will leapfrog Intel with their 7nm die shrink at a time Intel remains stuck on 14nm desktop CPU's that's sprinkled with security issues on top :eek:
Click to expand...

Yeah it is weird - ~2 years ago they were on the cusp of a 10nm line-up and then supposedly a ramp to a new product before the end of 2017 and now it isn't even assured they will make 2019. Its kind of funny as they set out to make what would have been a 3 year head start on the areas important to CPU performance with a 7nm node with their 10nm and by the time they actually bring products to market they will probably be chasing TSMC's second generation 7nm node which will have all those refinements they they may not even manage to achieve on their 10nm.

Intel's 14nm though is more like 10nm elsewhere and the ++ variant still has room for improvement (they probably didn't bother as they thought they'd be ramping to 10nm) with some changes to track and lithography they can probably get another 15-16% performance out of it - though that falls a long way short of the performance that can come from density at ~7nm.
 
That's what happens when you rest on your laurels. They are going to have to pull something out of the bag if they are going to catch AMD in years to come.
 
Oh, im sure i read that cannon lake was getting a built in fix. Maybe that meant a software one, not hardware?
 
NinjaCool said:
Semiaccurate claims that Intel’s 10nm process that’s coming out in late 2019 is now a cut down version that’s closer to a 12nm process so Intel can actually get something out to us :eek:

They also claim Intel is still facing major yield issues on their 10nm process.
https://www.semiaccurate.com/2018/08/02/intel-guts-10nm-to-get-it-out-the-door/
Click to expand...

I wouldn't be surprised. For the Q4 2019 we know it from the last Intel official announcement.
If that is now 12nm and not 10nm we shall see. But if it is, then they are going to be in trouble especially on the server market.
 
Panos I'm pretty sure Intel will still call it a 10nm process no matter what they do to it.
It might be hard for us to know if Intel ends up taking a shortcut or two but the Semiaccurate article seems quite revealing.
 
I forget the name of the company that uses electron microscopes to check out process nodes/new chips and generally do reports and give the basics on real feature size seen. After that story that site will make for an interesting read at some point in 2020. It also doesn't sound surprising, to go from 'hey we're launching' to, small dual core can't even work with the gpu enabled to 18 month (minimum now) delays certainly implies that yields were genuinely dreadful.

He states 60% is where most fabs want to be for early volume production yields that you then want to ramp up, you can do risk production at significantly lower yield particularly in two scenarios, it gives you a market leading advantage and/or you sell server chips with silly margins. To me it's certainly obviously Intel has to be way below 30-40% range that you could be selling small Xeons for and still be profitable if they gave huge power advantage. Hell Intel pushes loss leading stuff all the time with contra revenue in mobile and elsewhere, they could simply push chips at a loss simply to not face terrible news in the stock market with continued failure with 10nm so if it could be out at all... it would be imo. IT certainly implies the node is just simply not capable of producing fully working chips. Remember even with the stated 8-10% yield, that's for half working dies with the gpu disabled, that's shocking.

THe smallest feature that really matters on their node is the 36nm metal pitch and that is using cobalt which is the big new step for the node... so that will almost certainly be the part that probably gets cut. Back to copper, match TSMC/Global with 40mm metal pitch, I believe their fin pitch was already larger which probably puts them at a ~12nm node.

I mean, I've heard of year long delays for nodes, but we're talking 4 years from 2015 to when they state it will now come end of next year. This is the industry leader on nodes for a few decades.... 4 years is a truly major monumental screw up such that if it could have been fixed it's likely it would have been literally a couple of years ago already. There were already major signs of problems with their 14nm with big delays that for Intel were unheard of and 10nm just looks to have doubled down on their issues.

Other fabs have delays but it's usually more about timing or fabs not being ready. Glofo delayed/cancelled 32nm but that was more because the fab was simply finishing in a midpoint between 32nm and 28nm so it made little sense to be on 32nm. Then when the fab was finally finished it was their first fab bring up, it was the first new fab for anyone at AMD for well, a long long time so that isn't surprising. A 4 year delay is... we're talking 1-2 full process node steps that can happen in that period of time.


Coupled with that Charlie is also saying Icelake has a major performance flaw that won't be fixed because it's fundamental to the design.


EDIT:-

https://www.extremetech.com/computi...ed-reveals-impressive-finfets-13-layer-design

not that site but it links to a story and shows some pictures. Chipworks is the site that does scans of chips and shows off feature sizes in real world chips. Their first pics of 10nm chips will interesting to see for sure.
 
drunkenmaster said:
Their first pics of 10nm chips will interesting to see for sure.
Click to expand...

Yeah. Atm we know that Intel is trying to "flush" the system making a 10nm dual core CPU. However it shows how many issues have with the process because attached 10nm IGP is not working and is cut off due to manufacturing issues. (is the biggest block of transistors in those CPUs).
And a dual CPU at 10nm is pretty small and basic chip these days. Is not a 28-32 core monolithic server behemoth. And to design something similar to AMD (chiplets), they will have to start from scratch scrapping all the "Lakes" putting them behind another 4 years.
 
drunkenmaster said:
The smallest feature that really matters on their node is the 36nm metal pitch and that is using cobalt which is the big new step for the node... so that will almost certainly be the part that probably gets cut. Back to copper, match TSMC/Global with 40mm metal pitch, I believe their fin pitch was already larger which probably puts them at a ~12nm node.
Click to expand...
Dropping Cobalt seems likely and they could also remove things like Contact over Active Gate but changing too much would require a physical redesign.
If this is what is actually happening at Intel, then I think it's going to be hard to get this out before Christmas 2019.
 
I mean at this point anyone who straight up believes holiday season 2019 is crazy because Intel has promised chips for this process since 2015 and hell it was mid/late 2017 they 'launched' 10nm Cannonlake saying it would be available soon in mobile, not much longer after desktop. There is no way at that point they didn't realise they had to delay.

I'd be more surprised if 10(or 12nm) stuff came out end of next year than if it didn't.


Though if Icelake really has an issue as well it doesn't really matter is if it's got a flaw such that it's slower than Skylake, who is going to want it.

I think much like the delays with 14nm and Broadwell ending up launching on desktop around a month or two before Skylake, Icelake might only appear as a "see, we didn't lie to the shareholders/stockmarket/FCC" launch just as Broadwell.

There is also the appearance of a new server chip for 14nm that seems to slot in mighty close to the supposed 10nm Icelake server chips launch that also somewhat suggests another delay will be announced in the future. If there is another delay I think they'll announce it with Q1 results next year at the latest.
 
It strikes me that delays due to yield are only going to get worse as dies shrink. IBM have managed to create a 5nm chip, but I can't see it going much further. I mean, an individual atom of silicone is 0.2nm, we're right up against the limits of silicone transistors.

I'm looking forward to see what's next, carbon nanotube transistors perhaps or maybe even optical transistors? We shall see!
 
NinjaCool said:
It's quite shocking when you think about Intel initially planned to release their first 10nm CPU's in 2016!
Click to expand...
10nm was originally scheduled for late 2015... and 7nm for 2017.
https://www.nextbigfuture.com/2011/06/intel-roadmap-from-june-2011-with-7nm.html

With all the resources and talent Intel has can't think anything, but that Intel also has management issue.


Calo said:
How much of this delay is them having problems trying to close all their security holes though?
Click to expand...
Manufacturing node has nothing to do with security holes in architecture.
Would you blame it on material, if some safe box maker did such thin walled safe that you could cut it open with scissors?
 
You must log in or register to reply here.
Back
Top Bottom