Intel 10nm/7nm CPU/GPU Rumor Thread : Cannonlake, Icelake, Tigerlake, Sapphire Rapids, etc.

Um what are we looking at?
No idea. I was never able to read LTT forums, and the fact that English is not that guy's first language makes it even worse. I really don't think a low-tier support employee would be leaking stuff like this openly on a public forum, either.
 
10nm Icelake Xeon coming soon

DacQe61W4AAbTsf.jpg
 
The server market has much better financials than HPDT. Their focus is on money, not AMD.

Dead on here. Most of the growth in the industry is coming from the increase in gpu dependent systems for ai ect. They all need a powerful server cpu to run those gpus and you can believe these companies will be shopping around for the best buys for performance per watt, security (cough: specdown), overall performance and yes cost. All we get on HEDT is essentially trickle down tech from server R/D that can be re-purposed and fine tuned for the desktop. AMD is no different. Threadripper wouldn't exist without EPYC.
 
Told ya 10nm is a nightmare for them, they have to wait until the tech is there for it, they have wasted billions trying to get it done faster. They have focused so much energy on Fabs that it has come at a cost in other divisions.
 
Well anyways, I guess I should recap what BK said about 7nm last night.

It is forecasted to make use of EUV technology (10nm is the last one NOT to)

The have reduced density goals from 2.7x to 2.4x

There are "a lot of changes in the 7nm process" hinting at a large departure from 10nm(Hopefully not ass backwards use of cobalt everywhere)

Talked about EMIB/Mix match design, 7nm cores, 22/14nm IP.
 
tI6bxCD.png


ICL - Y supposedly upto 4c + GT2, LPDDR4, 3733mhz
ICL U supposedly upto 4c + GT2 (no surprise there) DDR4 3200

Looks like Chrisdar was right, at least on the BGA Socket Sizes as confirmed by Intel's own website:
DfQ7s7MWAAID1Ts.jpg

(he called it way back in Aug'17)
DfQ7dV7X4AEjUGG.jpg

Source

Intel's Website also confirmed many others:
BopfWCQ.png

(ICX SP, ICX-D)
AF6g2EN.png

(CNL Y, CNL U)

RspLulb.png

(WHL U)

fsdBsdJ.png

(CLX SP, CLX AP)

CNL-U: BGA1510

CNL-Y: BGA1392

WHL-U4(+)2: BGA1528

CLX-SP: LGA3647

CLX-AP: BGA5903

ICL-Y: BGA 1377

ICL-U: BGA 1526

ICL-SP: LGA 4189

ICL-D: BGA 2579
 
Looks like Chrisdar was right, at least on the BGA Socket Sizes as confirmed by Intel's own website:
View attachment 79920
(he called it way back in Aug'17)
View attachment 79921
Source

Intel's Website also confirmed many others:
View attachment 79922
(ICX SP, ICX-D)
View attachment 79923
(CNL Y, CNL U)

View attachment 79924
(WHL U)

View attachment 79925
(CLX SP, CLX AP)

CNL-U: BGA1510

CNL-Y: BGA1392

WHL-U4(+)2: BGA1528

CLX-SP: LGA3647

CLX-AP: BGA5903

ICL-Y: BGA 1377

ICL-U: BGA 1526

ICL-SP: LGA 4189

ICL-D: BGA 2579

ughh not ICL consumer part. ICL-SP with 4189 pins thats even bigger than current xeon socket.. hope 10nm+ is worth a lot of power reduction.
 

Well, mainstream Ice Lake will be nice for mobile CPU power consumption I'm sure, but for us overclocking desktop gamers, I'm expecting a mere incremental 5-10% "IPC" increase for relevant applications almost or entirely offset by lower max clock speeds versus highly matured 14nm++ parts. And octacore Coffee Lakes are supposedly happening so... whatever, unless Ice Lake is somehow an unprecedented jump we haven't seen the likes of since around Sandy Bridge.

Excluding the recent core count increases (which are certainly welcome to some degree in the consumer space), all indications are that we've still hit reasonable frequency limits and that high-performing architectural developments are exceedingly difficult at this point. Intel should give us eDRAM/L4 cache. That'll make for some quantified consumer-level gains (Seriously, do it; what are they waiting for? Do it and eradicate Zen's chance of gaming competitiveness.). Invested all that time and energy into making the tech, demoed it back in 2015 in Broadwell-C, and they still haven't used that little ace up their sleeve... <grumble>
 
Last edited:
Well, mainstream Ice Lake will be nice for mobile CPU power consumption I'm sure, but for us overclocking desktop gamers, I'm expecting a mere incremental 5-10% "IPC" increase for relevant applications almost or entirely offset by lower max clock speeds versus highly matured 14nm++ parts. And octacore Coffee Lakes are supposedly happening so... whatever, unless Ice Lake is somehow an unprecedented jump we haven't seen the likes of since around Sandy Bridge.

Excluding the recent core count increases (which are certainly welcome to some degree in the consumer space), all indications are that we've still hit reasonable frequency limits and that high-performing architectural developments are exceedingly difficult at this point. Intel should give us eDRAM/L4 cache. That'll make for some quantified consumer-level gains (Seriously, do it; what are they waiting for? Do it and eradicate Zen's chance of gaming competitiveness.). Invested all that time and energy into making the tech, demoed it back in 2015 in Broadwell-C, and they still haven't used that little ace up their sleeve... <grumble>

I am also expecting 5--10% IPC gains in serial code. Plus huge vector ~2x IPC increase due to AVX512. 10nm+ max performance could be worse than 14nm++ (e.g max 4.8GHz instead 5.0GHz) but I expect a nice increase in base clocks due to lower power consumption. Add also Meltdown and Spectre mitigation in silicon to help performance. People looking for higher clocks would wait to TigerLake and 10nm++.

I don't think L4 is a magic solution to the memory wall. Otherwise everyone would be adding more cache layers.
 
I am also expecting 5--10% IPC gains in serial code. Plus huge vector ~2x IPC increase due to AVX512. 10nm+ max performance could be worse than 14nm++ (e.g max 4.8GHz instead 5.0GHz) but I expect a nice increase in base clocks due to lower power consumption. Add also Meltdown and Spectre mitigation in silicon to help performance. People looking for higher clocks would wait to TigerLake and 10nm++.

I don't think L4 is a magic solution to the memory wall. Otherwise everyone would be adding more cache layers.

I mean, it'll have its limits and I'm sure cost is a factor in deciding whether to implement it. But Broadwell-C benefited nicely from it. If Intel are really moving towards separate dies for GPUs and CPUs especially, using the saved silicon budget to throw in some L4 on the higher-end consumer SKUs and gaining another 10-30% in even a small variety of memory-intensive games and stuff would look very nice in reviews. Source on the up-to-30% potential gain, not accounting for unknown variable changes since Broadwell such as current higher-end DDR4 speeds:

https://www.techpowerup.com/forums/...c-edrams-impact-on-gaming-performance.236514/


If I recall correctly, Haswell/Broadwell-E also seemed to gain some notable performance from their larger (pre-mesh) L3s rather than the cores at times. I'm convinced if there's anything left to substantially gain in performance (that Intel's designs in particular can effectively scale up), it'll involve more cache. Manufacturing cost versus whether Intel feel they need it atm being the only real limiting factors.
 
Last edited:

im assuming thats end of 2019 right? usually 2nd half. so sept to dec

Well, mainstream Ice Lake will be nice for mobile CPU power consumption I'm sure, but for us overclocking desktop gamers, I'm expecting a mere incremental 5-10% "IPC" increase for relevant applications almost or entirely offset by lower max clock speeds versus highly matured 14nm++ parts. And octacore Coffee Lakes are supposedly happening so... whatever, unless Ice Lake is somehow an unprecedented jump we haven't seen the likes of since around Sandy Bridge.

Excluding the recent core count increases (which are certainly welcome to some degree in the consumer space), all indications are that we've still hit reasonable frequency limits and that high-performing architectural developments are exceedingly difficult at this point. Intel should give us eDRAM/L4 cache. That'll make for some quantified consumer-level gains (Seriously, do it; what are they waiting for? Do it and eradicate Zen's chance of gaming competitiveness.). Invested all that time and energy into making the tech, demoed it back in 2015 in Broadwell-C, and they still haven't used that little ace up their sleeve... <grumble>

i think 5% IPC is too much of an expectation tbh, we still donno about how the "fixed" cpu wil perform. theres like 0 consumer software uses avx512, we'll be lucky if we can find 1 or 2 for avx2.
 
I wanted to bump this thread for pure amusement to many here. Reading the first few pages was all about "ICL Q218" this and "CNL has shipped" that leading many to believe all was on track.

Megatron Bonus - The 9700k may still be 14nm and just more skylake:
http://www.tomshardware.com/news/core-i7-9700k-intel-9th-generation,36826.html

Maybe this will be the rumored Whiskey Lake.
That Article is pure speculation - they know literally nothing, and no, Whiskey Lake is Mobile Only

I guess they knew a lot more than you gave them credit. Shintai must have known the writing was on the wall shortly before this since he has been MIA ever since.
HWInfo to add "improved support for Intel Icelake"
View attachment 84243
Link

And yet the most hardcore of super fans Dayman and Juangra continued to show "leaks" long after that post that all is still on track including the post above and others involving Server ICL-X by Q2 2018. Best thread ever LOL.
 
Last edited:
  • Like
Reactions: N4CR
like this
I wanted to bump this thread for pure amusement to many here. Reading the first few pages was all about "ICL Q218" this and "CNL has shipped" that leading many to believe all was on track.



I guess they knew a lot more than you gave them credit. Shintai must have known the writing was on the wall shortly before this since he has been MIA ever since.


And yet the most hardcore of super fans Dayman and Juangra continued to show "leaks" long after that post that all is still on track including the post above and others involving Server ICL-X by Q2 2018. Beth thread ever LOL.

I haven't went back to update the original post in a long time, those availability dates were guesses by my based on the ramp in H2'18, not a pushed back one in "sometime 2019" lol. Only time I come back here is to post about updates/leaks on Intel 10nm "products". Go create drama somewhere else.
 
I haven't went back to update the original post in a long time, those availability dates were guesses by my based on the ramp in H2'18, not a pushed back one in "sometime 2019" lol. Only time I come back here is to post about updates/leaks on Intel 10nm "products". Go create drama somewhere else.

What you did was blow off a legitimate leak from Tom's back in early April and the continued to trickle out leaks about ICL and CNL through 'microcode' all the way to the end of June.
 
And yet the most hardcore of super fans Dayman and Juangra continued to show "leaks" long after that post that all is still on track including the post above and others involving Server ICL-X by Q2 2018. Beth thread ever LOL.

I didn't even reply to your post about the TomsHardware article about the 9700k. Apparently for you I am guilty of posting a link to a slide from a presentation about forthcoming server platform. A slide that says "2018/2019" release date, not "Q2 2018". And I also posted a link to a tweet about desktop 10nm planned for 2019.
 
Last edited:
at the moment 10nm or 10nm+ dont matter guys. what matter is if this 8 core be any good and if it comes with 14nm+++
 

10nm Icelake Xeon coming soon

View attachment 65761

Apparently coming soon is Q3/Q4 2020. Launching prior to this is "Cooper Lake".
https://www.anandtech.com/show/1311...ap-leaks-cooper-lakesp-ice-lakesp-due-in-2020

And then there is this:
https://www.techpowerup.com/246306/intel-stuck-with-14nm-processors-till-holiday-2019

Paraphrasing - 10nm products will first arrive Holiday 2019. So does this mean available to most in 2020? What a mess!
 
Here let me help

"10nm is super f'n late and delayed again!"

Of course, it was originally planned for about 2015. The node transition was very aggresive and the plans failed miserably. We all know that. It is repeated in media and forums again and again.

What is not mentioned so often (in fact it is literally ignored) is that Glofo also planned 10nm for about 2015 and not only delayed it but canceled it twice (first canceled 10XM in 2016 then canceled 10LP about a year ago) and now goes for 7LP directly. And don't forget that what other foundries call 7nm is an inferior node to what Intel calls 10nm.
 
Of course, it was originally planned for about 2015. The node transition was very aggresive and the plans failed miserably. We all know that. It is repeated in media and forums again and again.

What is not mentioned so often (in fact it is literally ignored) is that Glofo also planned 10nm for about 2015 and not only delayed it but canceled it twice (first canceled 10XM in 2016 then canceled 10LP about a year ago) and now goes for 7LP directly. And don't forget that what other foundries call 7nm is an inferior node to what Intel calls 10nm.

Bingo. And to top it all off their current 14nm process will still be better in a lot of aspects, compared the competition. Intel sucks, but their process nodes don't...
 
Back
Top