Discussion Intel current and future Lakes & Rapids thread

[Eug]

I keep reading this as company's name as ICK! Knowledge instead of IC Knowledge.

IC Knowledge report: Technology and Cost Trends at Advanced Nodes

I just mention it since in that report, they tried to normalize the various node technologies from the various companies into a table for comparison.

 

[witeken]

I keep reading this as company's name as ICK! Knowledge instead of IC Knowledge.

IC Knowledge report: Technology and Cost Trends at Advanced Nodes

I just mention it since in that report, they tried to normalize the various node technologies from the various companies into a table for comparison.

My opinion is still that this "normalization" is a futile and misleading practice because it clouds the real metrics. Really, there is no reason at all not to use the direct metric that a node name supposedly has to capture: real world density.

 

[Dolan]

My opinion is still that this "normalization" is a futile and misleading practice because it clouds the real metrics. Really, there is no reason at all not to use the direct metric that a node name supposedly has to capture: real world density.

100% agree with you. Lets use real world density they use in real world applications which generates majority of revenue for them. For N14 it is 14mxtors Intel, 25m others. For 10 it will eventually be around 30m for Intel and 50 milion per square mm for others. Continuing 60vs100, 120vs200...

Or you don't agree about dates? Well then take 2019 when Intel maybe start risk production of 10nm, while other planing risk production of 5nm for same year. This gives 30m xtors for Intel versus 200m xtors for others in same time frame.

Well, now i realized, why you are so angry, and you had to accuse respected analysts / companies from misleading.

 

[jpiniero]

https://twitter.com/david_schor/status/998608222217891841

Ooh boy. David Schor seems to think that 10 nm is more realistically a "small ramp" at the end of 2019 and the larger ramp in 2020. IOW what Intel was saying earlier this year about 2018 and 2019. Have to think Icelake (both client and server) would be canned unless they do more low volume products with it and they will just roll with Sapphire Rapids and Tigerlake.

 

[mikk]

I really doubt Intel can know how the 10nm yield progresses over the next 12-18 months right now. And what is a small ramp for him by the way, I mean there is a first 10nm SKU in market with another upcoming CNL-Y, we may hear more at Computex. So there is a small 10nm ramp this year. Hard to believe Intel isn't going to increase the ramp till end of 2019.

 

[jpiniero]

I really doubt Intel can know how the 10nm yield progresses over the next 12-18 months right now. And what is a small ramp for him by the way, I mean there is a first 10nm SKU in market with another upcoming CNL-Y, we may hear more at Computex.

The implication is that yields are so bad that Intel is losing money selling the 10 nm chips, even if they sold it at typical i3 prices. So I imagine the volume is not much, only for test purposes. And they know things are so bad that it's going to take a long time to fix but would leave the door open if that magically changed.

But the die is so small even at really awful yield there might be enough for something like Surface, which is probably maybe only going to sell 50-100k/quarter.

 

[mikk]

12 months is a long time for me, 18 months even longer. I don't think Intel is able to predict the progress or yield rate for a timeframe in late 2019. It makes David Schor unreliable when he says this.

 

[Dayman1225]

https://twitter.com/david_schor/status/998608222217891841

Ooh boy. David Schor seems to think that 10 nm is more realistically a "small ramp" at the end of 2019 and the larger ramp in 2020. IOW what Intel was saying earlier this year about 2018 and 2019. Have to think Icelake (both client and server) would be canned unless they do more low volume products with it and they will just roll with Sapphire Rapids and Tigerlake.

Much more interesting details in that article past the paywall.

 

[Lodix]

The earliest you should expect a real product from intel manufactured at 10nm would be 3Q 2019. And that is 10nm not 10nm+.

 

[Dayman1225]

The earliest you should expect a real product from intel manufactured at 10nm would be 3Q 2019. And that is 10nm not 10nm+.

Then there is no real product to be had. 10nm is one and done on Cannonlake. (Unless you are referring to modems and other stuff)

 

[Lodix]

Then there is no real product to be had. 10nm is one and done on Cannonlake. (Unless you are referring to modems and other stuff)

They can use it for Icelake, but it will have lower clocks than planned. There is no point in releasing Cannolake CPUs in 2019 because it has the same architecture as Skylake.

10nm+ is an improvement and matured 10nm. They can't just jump from 14nm to 10nm+ without the learning curve of the yields and maturity of the process. Or they would have to delay it even more, but no point in that.

 

[LTC8K6]

The earliest you should expect a real product from intel manufactured at 10nm would be 3Q 2019. And that is 10nm not 10nm+.

If that is true, then Intel is entirely wasting it's time and money on 10nm.

 

[LTC8K6]

They can use it for Icelake, but it will have lower clocks than planned. There is no point in releasing Cannolake CPUs in 2019 because it has the same architecture as Skylake.

10nm+ is an improvement and matured 10nm. They can't just jump from 14nm to 10nm+ without the learning curve of the yields and maturity of the process.

Since they had a working 10nm chip in January of 2017, 10nm has to be somewhat mature by now.

 

[Lodix]

If that is true, then Intel is entirely wasting it's time and money on 10nm.

Why? They need the density and efficiency improvements it brings to compete with the rest of the market in different segments.

 

[Lodix]

Since they had a working 10nm chip in January of 2017, 10nm has to be somewhat mature by now.

Are you for real? They can't even manufacture at high volume a tiny mobile SOC with 2 cores CPU, the GPU totally disabled and no improvements in efficiency compared to 14nm. Because yields are terrible now, imagine 1 year before.

 

[Lodix]

Then there is no real product to be had. 10nm is one and done on Cannonlake. (Unless you are referring to modems and other stuff)

Just answered to your question to the other guy. They will use whatever is available to be able to compete with a market transitioning to Foundries 7nm.

 

[Dayman1225]

If that is true, then Intel is entirely wasting it's time and money on 10nm.

14nm can't last forever, unless you'd prefer 14nm CPUs until 7nm sometime 202x. Plus issues at 10nm help yield learning at future nodes and provide valuable insight to possible future issues and smaller nodes

 

[Dayman1225]

Just answered to your question to the other guy. They will use whatever is available to be able to compete with a market transitioning to Foundries 7nm.

That is the problem, ICL is designed for 10nm+. There isn't much available to release, unless they do more 14nm.

 

[Lodix]

That is the problem, ICL is designed for 10nm+. There isn't much available to release, unless they do more 14nm.

There isn't much difference between 10nm amd 10nm+ in design if any. They have more than a year to do a small tweak and get it done ( if they were smart they would have started before as they saw the yield problems ).

 

[.vodka]

Intel Kabylake Client platform Kabylake Client System (Intel KBL S DDR4 UDIMM EV CRB)

Genuine Intel(R) CPU 0000 @ 2.60GHz (8C 16T 2.59GHz, 8x 256kB L2, 16MB L3)


Hello there, 8C Coffeelake! or whatever lake you are

 

[jpiniero]

12 months is a long time for me, 18 months even longer. I don't think Intel is able to predict the progress or yield rate for a timeframe in late 2019.

It is a long time. I think it just shows how bad of a situation it is and that it's not just tweaking the machines. Probably also have to make physical changes to the chips, and that takes time to complete and validate.

 

[LTC8K6]

A while ago, the word was that Intel was skipping 10nm altogether and going with 10nm+.

That is, we'd see no 10nm chips.

 

[LTC8K6]

Well, Whiskey Lake/Coffee Lake R 14nm chips it is, then.

 

[Beemster]

Genuine Intel(R) CPU 0000 @ 2.60GHz (8C 16T 2.59GHz, 8x 256kB L2, 16MB L3)

....2.6GHz all core????? when AMD is at 3.7GHz all core on 2700X with a likely 2800X in reserve??................can't be that bad....must be a chipset issue......dam them Core cores are power hungry........like I've been pointing out for some time..........the more cores....the harder it bites them........so of course no 10nm SERVER parts till God knows when............got to get to the multichip approach and a new core design

Intel Kabylake Client platform Kabylake Client System (Intel KBL S DDR4 UDIMM EV CRB)

Genuine Intel(R) CPU 0000 @ 2.60GHz (8C 16T 2.59GHz, 8x 256kB L2, 16MB L3)


Hello there, 8C Coffeelake! or whatever lake you are

 

[LTC8K6]

Genuine Intel(R) CPU 0000 @ 2.60GHz (8C 16T 2.59GHz, 8x 256kB L2, 16MB L3)

....2.6GHz all core????? when AMD is at 3.7GHz all core on 2700X with a likely 2800X in reserve??................can't be that bad....must be a chipset issue......dam them Core cores are power hungry........like I've been pointing out for some time..........the more cores....the harder it bites them........so of course no 10nm SERVER parts till God knows when............got to get to the multichip approach and a new core design

It's likely going to be quite close to the 8700K frequencies. Probably the same single core turbo, and slightly lower all core turbo. Plus the extra cache. It should perform like the 8700K single core and pull nicely ahead of it multi core.