Discussion Intel current and future Lakes & Rapids thread

Page 42 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
There was a German(not sure) site I gave a link to in one of these threads that showed Sapphire Rapid coming for client, just later than for Server.

Seems more likely now that they are going to continue on milking Skylake for client... until they can't. Sapphire Rapids might still be good for games, but we'll have to see.


Yeah. He has to register it with the SEC. Thing is, you can drive a Mack truck through the loopholes. Maybe now Intel will be more open about the state of 10 nm now that BK has sold his shares.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Seems more likely now that they are going to continue on milking Skylake for client... until they can't. Sapphire Rapids might still be good for games, but we'll have to see.

Sapphire Rapids generation might be no bigger of a change than Skylake vs Skylake-X. That means continued use of mesh for Server and Ring for client.
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
Sapphire Rapids generation might be no bigger of a change than Skylake vs Skylake-X. That means continued use of mesh for Server and Ring for client.

It's rumored to be a much more radical change, on the level of Conroe at the very least. We'll have to see. That doesn't mean of course it would be suitable for anything other than server.

I think Icelake mainstream will be using the mesh btw, where the only difference between it and the server version would be the 'bolted on' AVX-512 unit(s) and L2. And then they would just continue to shrink it to 10++ (Tiger lake), 7 (Alder Lake) but only when it makes financial sense. Maybe work on the IGP until the scalable one is ready if that's what they are really doing.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
It's rumored to be a much more radical change, on the level of Conroe at the very least. We'll have to see. That doesn't mean of course it would be suitable for anything other than server.

No, what I mean is the difference between Server and Consumer wouldn't be greater than how it is with Skylake. Of course the core is supposed to be radical. It should introduce new concepts like the ones we saw with Sandy Bridge. And Core. And Netburst. And Pentium. Those big changes.

I think Icelake mainstream will be using the mesh btw, where the only difference between it and the server version would be the 'bolted on' AVX-512 unit(s) and L2.

I disagree on the mesh. It'll make it especially complicated to reach the frequencies consumer cores are required to. What about the 2-4 cores on the Y and U parts? They won't do a redesign just for them. Mesh is even a toss up even in the HEDT space. It's only with many, many cores like Skylake-SP where it starts showing its strength. So we're talking 20+ cores. I think the latency issue for Skylake-X is also because of the non-inclusive L2. The current system of small L2 and large L3 on a ringbus works very well for most of us.
 
  • Like
Reactions: pcp7

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136

Dayman1225

Golden Member
Aug 14, 2017
1,152
973
146
Interesting, but potentially irrelevant. As the article itself references, LLVM Clang already has switches for Cannonlake behavior, never mind that we'll be seeing virtually no Cannonlake product on the market outside of very limited release.

Unlike Cannonlake, I do expect to see Icelake components eventually, but it's a question of when . . .
My guess is H1'19 but it seems others on here think H2'19 which I don't think will happen tbh.
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
Who wants to bet Intel announces Cannonlake tomorrow? ::crickets::

I disagree on the mesh. It'll make it especially complicated to reach the frequencies consumer cores are required to. What about the 2-4 cores on the Y and U parts? They won't do a redesign just for them.

It's primarily to cut costs by consolidating chip development. Considering Intel is doing Skylake-SP parts for Xeon-D, power consumption at the low end should be fine assuming they drop the bolted on AVX-512 and L2.

They can even drop Atom now that the Phi is using Core.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
H1 for mobile, H2 for desktop? Or are we seeing mobile Icelake in Q3/Q4 2018?

Chances for Icelake anytime before Q1 2019 are pretty low at this point. Depending on the rumors, the first Icelake may be Icelake-SP in Q2/Q3.

It's primarily to cut costs by consolidating chip development.

They can even drop Atom now that the Phi is using Core.
You wish. :)
 
Mar 10, 2006
11,715
2,012
126
Chances for Icelake anytime before Q1 2019 are pretty low at this point. Depending on the rumors, the first Icelake may be Icelake-SP in Q2/Q3.

You wish. :)

jpiniero's product strategy is basically what AMD would dream for Intel to do -- try to shoehorn inappropriate products and technologies in a lot of markets in the name of "cost reduction."
 
Last edited:

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
Maybe I missed it but do we have infos about Lakefield? It must be a product with graphics because it is listed in the graphics driver or Media SDK documentation (recently added).

unbenanntb8qsy.png

https://github.com/Intel-Media-SDK/MediaSDK/blob/master/doc/mediasdk-man.pdf

If the list is accurate it might come between Icelake and Tigerlake, although if it's Atom based not necessarily. I wonder if this is Core or Atom based or something else.
 

Dayman1225

Golden Member
Aug 14, 2017
1,152
973
146
Maybe I missed it but do we have infos about Lakefield? It must be a product with graphics because it is listed in the graphics driver or Media SDK documentation (recently added).

unbenanntb8qsy.png

https://github.com/Intel-Media-SDK/MediaSDK/blob/master/doc/mediasdk-man.pdf

If the list is accurate it might come between Icelake and Tigerlake, although if it's Atom based not necessarily. I wonder if this is Core or Atom based or something else.


I have never, ever heard of lake field.

EDIT: Ashraf Eassa claims its Core/Atom, big.LITTLE style product for mobile, also says its based on 10nm


https://twitter.com/TMFChipFool/status/946816452836945932

https://twitter.com/TMFChipFool/status/946817821635424256
 
Last edited:

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
Chances for Icelake anytime before Q1 2019 are pretty low at this point. Depending on the rumors, the first Icelake may be Icelake-SP in Q2/Q3.

Huh. I thought the first Icelake out of the gate would be Icelake-U or Icelake-Y. Now I'm confuzzled again.

jpiniero's product strategy is basically what AMD would dream for Intel to do -- try to shoehorn inappropriate products and technologies in a lot of markets in the name of "cost reduction."

Are they even going to continue Atom development, though? Core is starting to render Atom redundant, outside of the fixed-function hardware that some of the latest Atoms have that Core doesn't necessarily have.

I have never, ever heard of lake field.

EDIT: Ashraf Eassa claims its Core/Atom, big.LITTLE style product for mobile, also says its based on 10nm

Whaaaaat? Intel doing big.LITTLE?!?!
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Apparently so, with Core/Atom, when I said mobile i was not referring to phones btw, mainly tablets, 2 in 1s, thin and lights etc. This is gonna be a pain in the ass for microsoft lol!

This please!

evolution.jpg



Huh. I thought the first Icelake out of the gate would be Icelake-U or Icelake-Y. Now I'm confuzzled again.

The key point is "depending on the rumors".

If they can get any Cannonlake-Y parts out couple of weeks now, chances of Icelake-Y being early Q1 is likely. Some think though Cannonlake-Y is Q2. It isn't helped by the pattern that an annual cycle is little more than 12 months, 13, if you want to fix a number to it.
 

Bouowmx

Golden Member
Nov 13, 2016
1,138
550
146
I can see: 1 big core (Ice Lake) + 4 little cores (Goldmont Plus, or successor) to boost the low-cost lineup's (ex. Celeron N4100) single-thread performance and responsiveness. Or, would that not make it low-cost anymore?
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
I can see: 1 big core (Ice Lake) + 4 little cores (Goldmont Plus, or successor) to boost the low-cost lineup's (ex. Celeron N4100) single-thread performance and responsiveness.

That could be it. It could be just one variation.

Hmm. If the trend continues where the pace of low-cost cores outpace those of bigger ones, it would make sense in terms of area and power to put multiple of the smaller ones to replace some of the bigger cores. It's due to diminishing returns.

It starts to make sense why A11 Bionic chip used by Apple uses 2 high performance cores, but 4 lower end cores.
 

Dayman1225

Golden Member
Aug 14, 2017
1,152
973
146
https://twitter.com/TMFChipFool/status/946974924115193856

Ashraf claims the Cannonlake model being put into production is indeed a dual core with IGP fused off. Max turbo of 2.2 Ghz and 15W. Not sure who would buy this.

Pretty much also confirms that 10nm is still a yield disaster at this point.

2.2ghz gives away it's the processes problem, but fusing of IGPU, could be fore yield like you mentioned, or Intel some how messed up GEN10 design to where it's unusable with 10nm(or some other issues) but yield is the most likely. This is pathetic.
 
Last edited:

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
Brian Krzanich would say Cannonlake is on track! The problem is that they can't even sell this product for the low budget market without iGPU, and then only two cores with max 2.2 Ghz lol. It looks like there is more behind the scenes beside yield issues. There must be some design issue, maybe in combination with 10nm. There is hope this design flaw will be solved for Icelake/Gen11.
 

witeken

Diamond Member
Dec 25, 2013
3,899
193
106
Brian Krzanich would say Cannonlake is on track! The problem is that they can't even sell this product for the low budget market without iGPU, and then only two cores with max 2.2 Ghz lol. It looks like there is more behind the scenes beside yield issues. There must be some design issue, maybe in combination with 10nm. There is hope this design flaw will be solved for Icelake/Gen11.
A75 achieves over 3.3ghz, and that's obviously made with the SoC flavor of 10nm.