[Anandtech] Intel's Architecture Day 2018

[PeterScott]

Intel's Architecture Day 2018: The Future of Core, Intel GPUs, 10nm, and Hybrid x86

I know some bits of this were getting some mention in other threads, but it seems like it was big enough to warrant it's own discussion.

Highlights for me were a new architecture (finally) Sunny cove, that looks to have potential (4-5 wide allocation) to deliver Intels first real IPC improvement in MANY years, and also what should be a nice improvement in the the IGP, and a look at Intels multichip solutions(Foveros) and Intel Big-little.

The Q&A was also somewhat interesting, like where they admit being node-locked on their designs really messed them up, and they will be more node agnostic in the future.

Overall, more actual news than we have seen from Intel in some time.

 

[Markfw]

So AMD iterating on Zen is great, but Intel building better cores over time is going to get them "walked over"? OK.

Tweaking a new arch a couple of times is one thing thing. Intel is now on its 10th iteration or Core. Thats a little much, don't you think ? Time for a new arch ?

 

[Arachnotronic]

Tweaking a new arch a couple of times is one thing thing. Intel is now on its 10th iteration or Core. Thats a little much, don't you think ? Time for a new arch ?

Even Zen was built substantially from elements lifted (and of course modified) from Bulldozer and Jaguar.

The problem with Intel's Core today is that Intel hasn't changed the Skylake microarchitecture at all since...Skylake. Kaby Lake, Coffee Lake, and all of the other refreshes/derivatives are based on the exact same core with no changes. Sunny Cove is a really big change from Skylake. Had that been what Intel released in 2017 or 2018 (instead of just using the same Skylake core over and over again because they didn't have the foresight to port Sunny Cove to 14nm for the 2017/2018 products), people would probably be really happy with the progress Intel made in Core.

 

[Yotsugi]

Even Zen was built substantially from elements lifted (and of course modified) from Bulldozer and Jaguar.

The problem with Intel's Core today is that Intel hasn't changed the Skylake microarchitecture at all since...Skylake. Kaby Lake, Coffee Lake, and all of the other refreshes/derivatives are based on the exact same core with no changes. Sunny Cove is a really big change from Skylake. Had that been what Intel released in 2017 or 2018 (instead of just using the same Skylake core over and over again because they didn't have the foresight to port Sunny Cove to 14nm for the 2017/2018 products), people would probably be really happy with the progress Intel made in Core.

This.
There's nothing wrong with iterations, AMD is busy iterating Zen, Apple has been iterating Cyclone, Intel iterates on Core.
You can make the very same statement for dGPU vendors.

 

[Dayman1225]

This.
There's nothing wrong with iterations, AMD is busy iterating Zen, Apple has been iterating Cyclone, Intel iterates on Core.
You can make the very same statement for dGPU vendors.

Exactly. Though I can see why people want more from Intel because Skylake lasted so long due to 10nm delays. However it appears they are learning from their [10nm] mistakes and are no longer coupling IP and process nodes together.

Now lets watch with some popcorn as AMD and Intel duke it out for the foreseeable future.

 

[HurleyBird]

Tweaking a new arch a couple of times is one thing thing. Intel is now on its 10th iteration or Core. Thats a little much, don't you think ? Time for a new arch ?

Or you could argue it's the third iteration of the architecture, beginning with Sandy, then moving to Haswell, and then ending up at Skylake.

Depends how you delineate things. Of course, a more reasonable accounting is probably somewhere between the two extremes.

 

[Yotsugi]

Or you could argue it's the third iteration of the architecture, beginning with Sandy, then moving to Haswell, and then ending up at Skylake.

You can technically trace the legacy all the way down to P6.
Doesn't really matter.
The problem is Intel shipping SKL for, like, 5 years in a row if we're counting CPL as Q1 2020 launch.

 

[PotatoWithEarsOnSide]

Unless the concept of iteration has changed since I was in school, your improvements/optimizations get smaller and smaller with each iteration. The implication being that the longer you iterate the same product, the more your performance stagnates over time; lowest hanging fruit etc.
What is needed is a new design method every set period of time, with predictable iterations and life cycles.

 

[Spartak]

Tweaking a new arch a couple of times is one thing thing. Intel is now on its 10th iteration or Core. Thats a little much, don't you think ? Time for a new arch ?

Not sure what you were reading but it cannot possible be the Anandtech article. Calling those massive changes 'tweaks' is inflammatory behaviour not worthy of a moderator.

Sunny Cove from what was shown is a fundamental redesign akin to Conroe and Sandy Bridge.

People are too much hung up on names.

Zen = new from the ground up!
Saphire Rapids = no idea what it is but this is intel's 'new' design!



Mod callouts and mod baiting are not allowed.
Read the rules:
This is rule 13.
http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=threads/anandtech-forum-guidelines.60552/


esquared
Anandtech Forum Director

 

[Yotsugi]

Unless the concept of iteration has changed since I was in school, your improvements/optimizations get smaller and smaller with each iteration. The implication being that the longer you iterate the same product, the more your performance stagnates over time; lowest hanging fruit etc.
What is needed is a new design method every set period of time, with predictable iterations and life cycles.

This is semis we're talking about.
You iterate unless you want a radical design changes like going from wide and reasonably deep OoO to hyperpipelined housefires, which Intel did in NetBurst.

Sunny Cove from what was shown is a fundamental redesign akin to Conroe and Sandy Bridge

+1 AGU +1 store port, some scheduling changes and whatnot are not massive like SB.
Conroe was a beefed up Yonah anyway.

 

[DrMrLordX]

So AMD iterating on Zen is great, but Intel building better cores over time is going to get them "walked over"? OK.

Two points:

1). AMD is outrunning Intel right now. So yeah, it's great. Too bad for AMD they started from such a poor position in 2016. AMD might cross up Intel in nearly every market with Zen2, which ought to show you how poorly Intel is doing iterating on Core. Core is seriously in need of a kick in the pants.
2). In case you didn't read my post, Apple is outrunning both AMD and Intel. They have reason to fear the A-series chips. The A12x is a 10W monster. At this rate, Apple will bury both of them. All Apple has to do is license their A-series cores to Cavium (or just buy Cavium) and they can attack the cellphone, tablet, laptop, and server market all in one shot. I don't see Apple buying Cavium though, since they've gone so far out of their way to deprecate all their server-based product lines.

 

[Yotsugi]

Two points:

1). AMD is outrunning Intel right now. So yeah, it's great. Too bad for AMD they started from such a poor position in 2016. AMD might cross up Intel in nearly every market with Zen2, which ought to show you how poorly Intel is doing iterating on Core. Core is seriously in need of a kick in the pants.
2). In case you didn't read my post, Apple is outrunning both AMD and Intel. They have reason to fear the A-series chips. The A12x is a 10W monster. At this rate, Apple will bury both of them. All Apple has to do is license their A-series cores to Cavium (or just buy Cavium) and they can attack the cellphone, tablet, laptop, and server market all in one shot. I don't see Apple buying Cavium though, since they've gone so far out of their way to deprecate all their server-based product lines.

I don't see Apple going for servers, or being able to cook a 60+ core dies either.
They make decent SoCs and so be it.

 

[krumme]

It seems you missed the news that Intel's Custom Foundy efforts are no more.

Were there ever any? You can't be a foundry if your process is tied to design and vice versa. Impossible.

 

[Yotsugi]

Were there ever any? You can't be a foundry if your process is tied to design and vice versa. Impossible.

Intel very much offeres SoC variants of their nodes plus 22FFL is/was a thing.

 

[krumme]

Intel very much offeres SoC variants of their nodes plus 22FFL is/was a thing.

Their entire process is differently geared. They were totally uncompetitive out the gate and their results matched it.
3 and 5nm needs to be fonded. How is that done?

 

[Arachnotronic]

Were there ever any? You can't be a foundry if your process is tied to design and vice versa. Impossible.

I think you might be misunderstanding what Intel means by decoupling IPs from process.

 

[krumme]

I think you might be misunderstanding what Intel means by decoupling IPs from process.

I know this is not what was said. Obviously they can't go out and say what I write. I am trying to go a step further along the same thinking and development.

 

[Arachnotronic]

I know this is not what was said. Obviously they can't go out and say what I write. I am trying to go a step further along the same thinking and development.

I think you are misunderstanding. The decoupling of process and IP isn't really a statement around process -- it's a statement around the company's product development methodology.

 

[krumme]

I think you are misunderstanding. The decoupling of process and IP isn't really a statement around process -- it's a statement around the company's product development methodology.

Ok so when a design isn't tied to a certain process?
It goes both ways. Yes?
The consequences is what is interesting. But explain what you mean by product development. I only read what was on AT.

 

[Yotsugi]

It goes both ways. Yes?

No.

 

[Markfw]

Not sure what you were reading but it cannot possible be the Anandtech article. Calling those massive changes 'tweaks' is inflammatory behaviour not worthy of a moderator.

Sunny Cove from what was shown is a fundamental redesign akin to Conroe and Sandy Bridge.

People are too much hung up on names.

Zen = new from the ground up!
Saphire Rapids = no idea what it is but this is intel's 'new' design!

9900k is 9th gen. Not sure what 10th gen will be, but Sunny Cove is probably later than that, And until its a product, all the advertising in the world does not make it a product.

 

[Yotsugi]

9900k is 9th gen. Not sure what 10th gen will be, but Sunny Cove is probably later than that, And until its a product, all the advertising in the world does not make it a product.

10th gen is more Skylake.

 

[Markfw]

10th gen is more Skylake.

Thats what I thought. The callout about it being Sunny Cove is busted ??

 

[Yotsugi]

Thats what I thought.

I mean, it's Comet Lake + something.

The callout about it being Sunny Cove is busted ??

It may have some -Y and -U Icelake parts but I'm not currently betting on it.

 

[mikk]

I mean, it's Comet Lake + something.

At the moment it's not clear what Intel is planning for the next year or two!

 

[Yotsugi]

At the moment it's not clear what Intel is planning for the next year or two!

It pretty much depends on the state of 10nm.