Question Seems like the product stack get confusing when you are not leading

Hulk

Diamond Member
Oct 9, 1999
4,191
1,975
136
When Intel had a comfortable lead over AMD both in terms of architecture and process their product stack was pretty well organized. Sandy Bridge die shrink to Ivy Bridge, new architecture to Haswell, then Broadwell, etc... The tick/tock. But when their process stalled at 14nm things got very convoluted and confusing. Skylake core for 5 years and various process optimizations. Fail at 10nm, back to 14nm, ... All the while AMD was slowly but inexorably catching up...

I believe all of this craziness in naming schemes, processes, and architectures for specific markets such at Coffee Lake 1 and Coffee Lake 2, Kaby, Kaby R, etc... is a result of let's not say mad scrambling to get out competitive products in niches, but "highly panicked scrambling." When you are in the lead you can release parts when you want in a nice, orderly fashion. When you are behind you have to get something out the door to compete. It's kind of like how they say a boxer has a plan until the first time they get punched in the face.

And it looks like that is what's going on today. That's what I'm reading into this minor informational release on Rocket Lake.

You would think that Rocket Lake's core (Cyprus Lake) would be a derivative of their current "best" core, which seems to be Willow Cove. But it appears as though Cyprus Cove is a derivative (says Intel) of Sunny Cove with the addition of the new AVX 512 instructions, which are already present in Willow Cove? Does Cyprus Cove include the larger L1/L2 of Willow Cove?

I'm wondering.... Zen 3 is coming and Intel knows this. Based on what we know so far it looks as though Zen 3 will have an advantage over Sunny/Willow/Cyprus Cove when it comes to IPC in most areas. BUT, perhaps Intel is thinking they may be able to make up that difference in IPC with high clocks, which they can't do on their 10nm process. So they are taking the "Cove" architecture, stripping out anything that may restrict clock speed and creating Cyprus Cove at 14nm in hopes of possibly beating AMD in a few single threaded benches/apps on a core that is nearly as good IPC-wise as Zen 3 but can clock 10 or 15% higher. Hence we are hearing about this Cyprus Cove core, which seems to be a strange child of Sunny Cove and Willow Cove and manufactured on a very mature 14nm process... Viola! Rocket Lake S. And they may be able to carve out a couple more small market niches from Zen 3.

I am really enjoying the CPU Wars these days.
 

moinmoin

Diamond Member
Jun 1, 2017
4,933
7,618
136
That got to do not so much with leading or not but with the mess Intel created with its failing process node cadence. As you write yourself Intel had the tick/tock cadence when everything worked fine. The latest nodes though take too long to reach sufficiently decent yield so they keep running the old node as well, and put a mix of both in their newly launched product range. Even if Intel were still leading that'd look like the mess it is.
 

Hulk

Diamond Member
Oct 9, 1999
4,191
1,975
136
Possibly. But I'm thinking that without the pressure to make a release in response to AMD they would have remained on their own more orderly release cadence.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Possibly. But I'm thinking that without the pressure to make a release in response to AMD they would have remained on their own more orderly release cadence.

Actually it was during times of AMD's weaknesses where they failed to release products in an orderly cadence.

Kabylake came out early 2017. It should have been Cannonlake instead.
 

Hulk

Diamond Member
Oct 9, 1999
4,191
1,975
136
I think it's fair to say that the release of the original Zen CPU in early 2017 was the beginning of the pressure that has been building to date.

From 2017 to date is sure looks like Intel has been in a mad rush to push out the door whatever parts they felt would be competitive. Here are their releases from 2017 to date. That's 9 products in 3 years.

Kaby Lake R August 2017
Coffee Lake October 2017
Canon Lake (failed release) May 2018
Whiskey Lake August 2018
Amber Lake August 2018
Coffee Lake (again) October 2018
Comet Lake August 2019
Ice Lake September 2019
Tiger Lake September 2020

If we move backwards in time 9 product cycles from 2017 we have the following parts which takes us back 7 years to January of 2010.
Kaby Lake
Skylake
Broadwell
Haswell
Ivy Bridge
Sandy Bridge
Arrandale
Clarksdale
Gulftown

So pre Zen Intel released 9 parts over 7 years and post Zen 9 parts over 3 years. That's over double the product release rate right?
 
  • Like
Reactions: Tlh97

jpiniero

Lifer
Oct 1, 2010
14,509
5,159
136
AMD has nothing to do with it. OEMs demand new products every year, and with the 10 nm debacle they needed new 14 nm products to cover for it.

What they changed with each Skylake refresh may be due to AMD.
 

ondma

Platinum Member
Mar 18, 2018
2,718
1,278
136
When Intel had a comfortable lead over AMD both in terms of architecture and process their product stack was pretty well organized. Sandy Bridge die shrink to Ivy Bridge, new architecture to Haswell, then Broadwell, etc... The tick/tock. But when their process stalled at 14nm things got very convoluted and confusing. Skylake core for 5 years and various process optimizations. Fail at 10nm, back to 14nm, ... All the while AMD was slowly but inexorably catching up...

I believe all of this craziness in naming schemes, processes, and architectures for specific markets such at Coffee Lake 1 and Coffee Lake 2, Kaby, Kaby R, etc... is a result of let's not say mad scrambling to get out competitive products in niches, but "highly panicked scrambling." When you are in the lead you can release parts when you want in a nice, orderly fashion. When you are behind you have to get something out the door to compete. It's kind of like how they say a boxer has a plan until the first time they get punched in the face.

And it looks like that is what's going on today. That's what I'm reading into this minor informational release on Rocket Lake.

You would think that Rocket Lake's core (Cyprus Lake) would be a derivative of their current "best" core, which seems to be Willow Cove. But it appears as though Cyprus Cove is a derivative (says Intel) of Sunny Cove with the addition of the new AVX 512 instructions, which are already present in Willow Cove? Does Cyprus Cove include the larger L1/L2 of Willow Cove?

I'm wondering.... Zen 3 is coming and Intel knows this. Based on what we know so far it looks as though Zen 3 will have an advantage over Sunny/Willow/Cyprus Cove when it comes to IPC in most areas. BUT, perhaps Intel is thinking they may be able to make up that difference in IPC with high clocks, which they can't do on their 10nm process. So they are taking the "Cove" architecture, stripping out anything that may restrict clock speed and creating Cyprus Cove at 14nm in hopes of possibly beating AMD in a few single threaded benches/apps on a core that is nearly as good IPC-wise as Zen 3 but can clock 10 or 15% higher. Hence we are hearing about this Cyprus Cove core, which seems to be a strange child of Sunny Cove and Willow Cove and manufactured on a very mature 14nm process... Viola! Rocket Lake S. And they may be able to carve out a couple more small market niches from Zen 3.

I am really enjoying the CPU Wars these days.
Tiger Lake has already shown ability to reach 4.8 ghz in a power constrained environment on 10 nm SF. I think the use of 14nm for Rocket Lake is more related to yield problems with 10nm and wanting to prioritize laptop chips.