Discussion Intel current and future Lakes & Rapids thread

Page 22 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

krumme

Diamond Member
Oct 9, 2009
5,936
1,553
136
Okey so they can show a test wafer with arm75 with better density than 14nm samsung/tsmc. No shit. I have had s7 in hand for nearly 1½ year.
Its just embarrasing. I wouldnt be surpriced if the rest of the pack is far further with 7nm that have more or less the same specs on paper as Intel 10nm.
This show looks like the old GF process ppt back 5-6 years promising all kinds of magic. Now Intel do the information crap as everyone else. Not worth a read imo. Waste of time for my part. Will stay away until products hit shelves in meaningfull numbers, then i will look at performance.
 

Dayman1225

Golden Member
Aug 14, 2017
1,075
767
146
"They were claiming over 3.3GHz for A75 at 250uW(microwatts) per MHz" No idea if this is true, I found it in the comments of the Anandtech article posted by Ian, but that's below a watt if true. 0.825w to be exact.
 

Dayman1225

Golden Member
Aug 14, 2017
1,075
767
146
Okey so they can show a test wafer with arm75 with better density than 14nm samsung/tsmc. No shit. I have had s7 in hand for nearly 1½ year.
Its just embarrasing. I wouldnt be surpriced if the rest of the pack is far further with 7nm that have more or less the same specs on paper as Intel 10nm.
This show looks like the old GF process ppt back 5-6 years promising all kinds of magic. Now Intel do the information crap as everyone else. Not worth a read imo. Waste of time for my part. Will stay away until products hit shelves in meaningfull numbers, then i will look at performance.

Jeez, Intel hit a nerve.
 

Bouowmx

Golden Member
Nov 13, 2016
1,138
550
146
"They were claiming over 3.3GHz for A75 at 250uW(microwatts) per MHz" No idea if this is true, I found it in the comments of the Anandtech article posted by Ian, but that's below a watt if true. 0.825w to be exact.
Those numbers are in the MyDrivers article, though the Intel News room release says just > 3.0 GHz.
 
  • Like
Reactions: Dayman1225

IntelUser2000

Elite Member
Oct 14, 2003
8,076
2,879
136
Jeez, Intel hit a nerve.
I agree with krumme. It's all marketing jargon. Now marketing is a necessity for companies. It's fine, until they start pulling this nonsense.

Huawei Kirin 970 achieves 55 mil tr/mm2 in a product form. Intel claims TSMC has a density of 50 mil tr/mm2. That is a fantastic achievement. With Intel if you want to see a real world example that meets that criteria you have to dig deep. Like in SRAMs, where with Skylake cores you see it taking a miniscule amount of space(indicating it isn't even being used en masse). They had to cut L3 in Skylake-SP as well. The only part of the core where their density advantage starts to show strength they have to cut it.

I'll tell you more examples of theoretical advantage not translating into actual products. They claimed with 14nm Airmont core, the density advantage over a 22nm Silvermont one was 2.8x. Despite that, they went from a ~105mm2 die on Bay Trail to ~85mm2 on Cherry Trail.

With 14nm Knights Landing, it has a die size greater than 650mm2. Intel said it has 8 billion transistors. The density then is ~12 mil tr/mm2, far short of 28 mil tr/mm2 they claim on that slide.

Now I can tell you why they might have to do this. Their cores are usually reserved for very high clocks, and in that case the densest transistors can't always be used. But considering how well Apple is doing, and even smaller companies like Huawei, its a far-fetched excuse.

So you can meet Intel's density claims if you end up designing a chip with 90% SRAMs(cache) by area, or slow, simple chips. Great.
 

krumme

Diamond Member
Oct 9, 2009
5,936
1,553
136
If i was to derive anything from this its that:
1. They choose to display an arm cpu. And not only own x86 products or some sram chip or the usual stuff we see.
2. Using ‘industry standard design flows’

If anything it shows something about their intentions. Or at least they want us to beliewe that. And if true its a change in their -until now - consistent unsuccessfull foundry strategy in getting other customers. They still bet on it and aparently now more so. Imo its also the only way forward.
 

jpiniero

Lifer
Oct 1, 2010
11,322
3,052
136
BTW, I don't (don't think anyway) that Intel would keep the desktop on Coffee Lake for 2018 because the fMax is lower on Icelake. It has to be about yield.
 

Kaloi48

Member
Jun 2, 2016
30
27
91
It looks like 10nm Cannonlake has been delayed again.

Intel reschedules Cannon Lake launch to year-end 2018, say sources
http://www.digitimes.com/news/a20170920PD207.html
Intel has reportedly rescheduled the release of its next-generation Cannon Lake-based processors to the end of 2018, which has already affected notebook brand vendors' new projects and their suppliers, according to sources from the upstream supply chain...
 

jpiniero

Lifer
Oct 1, 2010
11,322
3,052
136
Intel reschedules Cannon Lake launch to year-end 2018, say sources
http://www.digitimes.com/news/a20170920PD207.html
Intel has reportedly rescheduled the release of its next-generation Cannon Lake-based processors to the end of 2018, which has already affected notebook brand vendors' new projects and their suppliers, according to sources from the upstream supply chain...
I think if that were to be true Intel would just cancel Cannonlake.
 

krumme

Diamond Member
Oct 9, 2009
5,936
1,553
136
Yep. Intel is adapting Capex - meaning reducing cost for years now. Fine it means delays but so be it. But its not a viable long term strategy.
One can question the integration benefit when Intel themselves mention industry standard design flows.
It makes sense now more than ever to look at the total separation of business unit here.
Perhaps we can see these arm75 baby steps as the first preparation and small excercises in that long term transition?
 
  • Like
Reactions: VirtualLarry

IntelUser2000

Elite Member
Oct 14, 2003
8,076
2,879
136
Which the chips you are comparing Intels chips to are exactly.
Not always either. Remember Atoms barely competed despite the disadvantage in density.

By making the general claim that their advantage is density it makes everyone else assume by going Intel they'd get all the advantages of historical Intel plus the advantage in density. But probably no-one uses their process as a foundry because if you start talking to them seriously everything falls apart.

I think if that were to be true Intel would just cancel Cannonlake.
It won't work. Icelake then would have to be the pipe-cleaner and they'd need to spec the part down if they don't want to have issues. "Pipe cleaner" chips are relatively simple so they can fix and allow the process to mature. You need to learn to crawl before you learn to walk, so to speak.

Intel reschedules Cannon Lake launch to year-end 2018, say sources
http://www.digitimes.com/news/a20170920PD207.html
This is hard to swallow. If its true the only advantage they ever had of being a chip vendor disappears. It means an 8-core 14nm Coffeelake in H2 2018 might be true.
 
  • Like
Reactions: raghu78

Dayman1225

Golden Member
Aug 14, 2017
1,075
767
146
This doesn't mean Icelake is necessarily delayed, as it's a "different" process. I don't care for cannonlake as long as Icelake is on track..
 

raghu78

Diamond Member
Aug 23, 2012
4,093
1,475
136
Not always either. Remember Atoms barely competed despite the disadvantage in density.

By making the general claim that their advantage is density it makes everyone else assume by going Intel they'd get all the advantages of historical Intel plus the advantage in density. But probably no-one uses their process as a foundry because if you start talking to them seriously everything falls apart.
Intel makes ridiculous claims in terms of transistor density inspite of real world product comparisons showing no such advantage. Intel Atom chips were low power mobile sub 2.5 Ghz chips just like Apple Ax series but they never beat them out in transistors/sq mm. Intel does not release transistor counts on Atom chips so as to avoid actual comparisons against chips made by TSMC or other foundries.

https://www.anandtech.com/show/7314/intel-baytrail-preview-intel-atom-z3770-tested

"Intel isn’t disclosing die size or transistor counts, which is ironic (and disappointing) given that Apple just disclosed both (or at least relative magnitude of one) for its A7 SoC"

Intel's density advantages are only in presentations and do not appear in real world product comparisons.

It won't work. Icelake then would have to be the pipe-cleaner and they'd need to spec the part down if they don't want to have issues. "Pipe cleaner" chips are relatively simple so they can fix and allow the process to mature. You need to learn to crawl before you learn to walk, so to speak.

This is hard to swallow. If its true the only advantage they ever had of being a chip vendor disappears. It means an 8-core 14nm Coffeelake in H2 2018 might be true.
I would like to quote charlie who has had his fair share of blunders but also fair share of accurate information well before it became evident to the public. charlie has been telling that Intel's 10nm process is basically broken.

https://semiaccurate.com/2016/12/22/intel-10nm-cannon-lake-back/
https://semiaccurate.com/2016/12/28/coffee-lake-says-dire-things-intels-10nm-problems/
https://semiaccurate.com/2017/09/06/intel-foundry-customer-bails/
https://www.semiaccurate.com/2017/09/11/semiaccurate-digs-intels-10nm-process-problems/

The problems with Intel are now beyond technological and extend to management. I am astonished that Intel messed up so badly from a position of leadership in semiconductor manufacturing. If Cannonlake is pushed to end 2018 we will see an Apple A12 chip built at TSMC N7 ship before Intel ships their first 10nm chip. TSMC will also ship higher volume of N7 chips in 2018 than Intel does on their 10nm. That must be hugely embarassing.
 

Bouowmx

Golden Member
Nov 13, 2016
1,138
550
146
Ice Lake being available "shortly after" Cannon Lake -> interpret to be similar to Skylake being available a quarter after Broadwell?
 

mikk

Diamond Member
May 15, 2012
3,461
1,202
136
It looks like 10nm Cannonlake has been delayed again.

Intel reschedules Cannon Lake launch to year-end 2018, say sources
http://www.digitimes.com/news/a20170920PD207.html
Intel has reportedly rescheduled the release of its next-generation Cannon Lake-based processors to the end of 2018, which has already affected notebook brand vendors' new projects and their suppliers, according to sources from the upstream supply chain...

I guess Brian Krzanich still claims Cannonlake is on track :rolleyes:

They should cancel Cannonlake completely. Year end 2018 means this would be a 2019 product if it comes. In year 2019 such Dualcore SKU even for 15W doesn't make sense. I would prefer KBL 4C and if I need more GPU horsepower go with Nvidias MX150.
 
  • Like
Reactions: Arachnotronic

Dayman1225

Golden Member
Aug 14, 2017
1,075
767
146
A rumour is a rumour until its confirmed. We'll see by the end of the year and early 2018 if this is true.
 

ASK THE COMMUNITY