What happened to Intel's Core roadmap during the 10nm debacle?

[Xebec]

The pain in Intel's transition from 14nm to 10nm is pretty well documented, but I was wondering what were Intel's original plans for Desktop and Server architectures before 10nm started to slip? (What cores were we supposed to have, besides Cannon Lake - what was removed from the roadmap, etc.)

...
Actually launched:

2015 - 14nm Skylake, ~ on time after Broadwell delayed
2016 - 14nm+ Kaby Lake (but was supposed to be 10nm Cannonlake)
2017 - *14nm++ Coffee Lake 6 core, 14nm Whiskey Lake for Mobile
2018 - 14nm++ Coffee Lake 8 core, and 1 SKU only of Cannon Lake for Chinese market (heavily disabled, low clocks)
2019 - 14nm Comet Lake and 10nm Ice Lake (mobile only - relatively low clocks and volume)
2020 - 14nm Rocket Lake and 10nm Tiger Lake (limited volume, but healthy frequencies)
2021 - 10nm Alder Lake, 10nm Ice Lake Servers
2022 - 10nm Raptor Lake
2023 - 10nm Raptor Lake Refresh, "4nm" Meteor Lake (somewhat limited volume, originally labeled "7nm"), Sapphire Rapids (launched 2 years late?)
2024 - (Arrow Lake)
2025 - (Arrow Lake Refresh)

*Node performance was substantially improved over base 14nm

Was the Desktop Roadmap supposed to be:

2016 - ***10nm Cannon Lake (up to 8 cores on desktop)
2017 - 10nm Ice Lake
2018 - Tiger Lake??
2019 - Meteor Lake ??
2020 - Alder Lake on "7nm"?
2021 - **Sapphire Rapids on 7nm?

**An Intel investor meeting in 2019 said Sapphire Rapids would be a 2021 launch.
*** Cannon Lake mentioned as a 2015 production target: https://web.archive.org/web/2018021...om/show/12436/intel-10nm-dualcore-cannon-lake Core count: https://www.tomshardware.com/news/intels-unreleased-10nm-cannon-lake-cpus-emerge

 

[tomatoma]

Personally, I wish I could have introduced Intel4 (7nm generation by other companies) early...
intel The Intel 10nm itself has a fairly good performance.
Although it is said to be the 10nm generation, it is competing with the processes of other companies' 7nm generation.
The problem is that Intel has increased the required performance of 10nm too much, which also leads to delays

 

[semiman]

When Cannon Lake messed up, everything started to wrong. Cannon lake wasn't supposed to simple die shrink, but with new architecture as well(a.k.a Palm Cove). These cove architectures are very foundation of Intel P cores today, ranging from laptops to large server CPUs(to Emerald Rapids). This left long lasting effects to all Intel Products' offerings.

 

[eding42]

When Cannon Lake messed up, everything started to wrong. Cannon lake wasn't supposed to simple die shrink, but with new architecture as well(a.k.a Palm Cove). These cove architectures are very foundation of Intel P cores today, ranging from laptops to large server CPUs(to Emerald Rapids). This left long lasting effects to all Intel Products' offerings.

Wasn’t Palm Cove just die-shrunk Kaby Lake with AVX-512 and a few tweaks?

Maybe you’re thinking of sunny cove?

 

[siliconbruh999]

Pal COve was a Tick not a Tock

 

[Daniel Nenni]

ChatGPT: What happened to Intel's Core roadmap during the 10nm debacle?

Short version: Intel’s client “Core” roadmap stalled and then detoured because 10 nm wasn’t ready in volume or at desktop-class clocks. Intel stretched 14 nm for years (multiple Skylake refreshes), limited 10 nm to a tiny “Cannon Lake” pilot, launched 10 nm first on laptops (“Ice Lake,” then “Tiger Lake”), and finally brought the new core to desktops by backporting it to 14 nm (“Rocket Lake”). The real reset arrived with Alder Lake on Intel 7 (the re-named 10 nm Enhanced SuperFin).

What changed, chronologically​

• 2016–2020: 14 nm extended (“PAO” era). Intel formally replaced Tick-Tock with Process–Architecture–Optimization, enabling multiple 14 nm waves: Kaby Lake (2016/17), Coffee Lake (2017, first 6-core mainstream desktop), and Comet Lake (2020, up to 10 cores).
• 
  
• 2018: 10 nm “Cannon Lake” barely shipped. Only a single low-volume SKU (Core i3-8121U) appeared, even lacking an active iGPU—effectively a proof-of-life, not a platform transition.
• 
  
• 2018–2019: 10 nm delays force a 14 nm bulge. Intel pushed high-volume 10 nm to 2019 and expanded 14 nm output to meet demand; the shift contributed to well-reported 14 nm CPU tightness.
• 
  
• 2019: 10 nm arrives on mobile first. Ice Lake (Sunny Cove) laptop parts began shipping mid-2019; desktops stayed on 14 nm.
• 2020: 10 nm SuperFin = Tiger Lake. Intel introduced 10 nm SuperFin with Tiger Lake, calling it the largest intra-node perf jump in its history. (Still laptop-first.)
• 
  
• 2021: Desktop gets a backport. Rocket Lake uses Cypress Cove—a Sunny Cove core backported to 14 nm—to finally move desktop IPC forward (with PCIe 4.0), even as core counts dipped from Comet Lake’s 10C to 8C.
• 
  
• Late 2021: Roadmap reset. Alder Lake launched on Intel 7 (renamed 10 nm Enhanced SuperFin) with a new hybrid P-core/E-core design, finally moving mainstream desktop off 14 nm.

The net effect on “Core”​

• Years of Skylake-derivatives on desktop (2015→2020) while 10 nm matured on mobile.
• 
  
• Cannon Lake plans collapsed into a token release; Ice Lake desktop was skipped, and Rocket Lake delivered the Sunny->Cypress Cove uplift via a 14 nm backport.
• 
  
• Naming divergence and confusion (10th-gen had both 14 nm Comet Lake and 10 nm Ice Lake), until Intel 7/Alder Lake unified the client roadmap again.

 

[Xebec]

ChatGPT: What happened to Intel's Core roadmap during the 10nm debacle?

TIL ChatGPT is a Semi expert (Sorry I couldn't resist )

The net effect on “Core”

• Years of Skylake-derivatives on desktop (2015→2020) while 10 nm matured on mobile. AnandTech ForumsWikipedia
• Cannon Lake plans collapsed into a token release; Ice Lake desktop was skipped, and Rocket Lake delivered the Sunny->Cypress Cove uplift via a 14 nm backport. WikipediaIntel Corporation
• Naming divergence and confusion (10th-gen had both 14 nm Comet Lake and 10 nm Ice Lake), until Intel 7/Alder Lake unified the client roadmap again. WIREDIntel Corporation

I'm curious if there any unnannounced cores that were canned or changed from the original plans? If Intel had continued tick tock, was this the intended roadmap?

2014 - Broadwell (tick) - 14nm
2015 - Skylake (tock)
2016 - Cannonalke (tick) - 10nm
2017 - Icelake (tock)
2018 - Tiger Lake (tick) - 7nm
2019 - Alder Lake (tock) (i.e. 2 years earlier than actual, on a better node)
2020 - Raptor Lake (tick) - 5nm

i.e. Was Alder Lake delayed by 2 years and put on an older process than originally planned? or were there other cores or processes planned?

P.S. There was another process roadmap that came out in 2019 (after 10nm was delayed a lot, but before the Swan and Gelsinger changes). This roadmap implied that nodes would overlap for a year unlike the prior Tick-Tock process.

 

[eding42]

TIL ChatGPT is a Semi expert (Sorry I couldn't resist )



I'm curious if there any unnannounced cores that were canned or changed from the original plans? If Intel had continued tick tock, was this the intended roadmap?

2014 - Broadwell (tick) - 14nm
2015 - Skylake (tock)
2016 - Cannonalke (tick) - 10nm
2017 - Icelake (tock)
2018 - Tiger Lake (tick) - 7nm
2019 - Alder Lake (tock) (i.e. 2 years earlier than actual, on a better node)
2020 - Raptor Lake (tick) - 5nm

i.e. Was Alder Lake delayed by 2 years and put on an older process than originally planned? or were there other cores or processes planned?

P.S. There was another process roadmap that came out in 2019 (after 10nm was delayed a lot, but before the Swan and Gelsinger changes). This roadmap implied that nodes would overlap for a year unlike the prior Tick-Tock process.

View attachment 3557

I think your timeline is more or less true up to alder lake or raptor lake. I'm guessing Intel only had a vague idea of what they wanted to do with Alder/Raptor and didn't define any specific nodes etc. They probably knew they wanted the hybrid architecture but that was about it.

But yeah apparently Alder Lake originally was supposed to come out in like 2020... would've crushed Zen 2 if that happened.

Raptor Lake is such a weird product because there's elements of desperation in the design (i.e. Intel pushing the ring bus to 4.8 GHz+ and causing degradation, increasing L2 size by another half meg per core, increasing core count again etc.) Apparently it was rushed out to beat Zen 4 to the market. I really don't know if that was Intel's original plan LOL.

 

[Xebec]

I think your timeline is more or less true up to alder lake or raptor lake. I'm guessing Intel only had a vague idea of what they wanted to do with Alder/Raptor and didn't define any specific nodes etc. They probably knew they wanted the hybrid architecture but that was about it.

But yeah apparently Alder Lake originally was supposed to come out in like 2020... would've crushed Zen 2 if that happened.

Raptor Lake is such a weird product because there's elements of desperation in the design (i.e. Intel pushing the ring bus to 4.8 GHz+ and causing degradation, increasing L2 size by another half meg per core, increasing core count again etc.) Apparently it was rushed out to beat Zen 4 to the market. I really don't know if that was Intel's original plan LOL.

So I guess the take-away for the roadmap change is that all products got delayed 2 years (meaning we're a full architecture behind where we should be today), rather than an architecture being skipped.

..

re: Raptor Lake - it's funny you say this. It impressed me in a way that Pentium 3 at 1.13 GHz on 180nm did. "Wow, Intel came out with something much higher performance on 180nm than I was expecting - it's even significantly beating Athlon K7 in benchmarks", and then it was recalled because it wasn't fully stable except under perfect conditions. Raptor Lake initially was very impressive because it completely blunted Zen 4 in application performance (at similar tiers but with high power draw) like Alder Lake vs Zen 3.

Raptor Lake is perhaps worse because they even doubled down a year later with 14th gen pushing clocks yet higher again, increasing stability issues.

 

[tomatoma]

So I guess the take-away for the roadmap change is that all products got delayed 2 years (meaning we're a full architecture behind where we should be today), rather than an architecture being skipped.

..

re: Raptor Lake - it's funny you say this. It impressed me in a way that Pentium 3 at 1.13 GHz on 180nm did. "Wow, Intel came out with something much higher performance on 180nm than I was expecting - it's even significantly beating Athlon K7 in benchmarks", and then it was recalled because it wasn't fully stable except under perfect conditions. Raptor Lake initially was very impressive because it completely blunted Zen 4 in application performance (at similar tiers but with high power draw) like Alder Lake vs Zen 3.

Raptor Lake is perhaps worse because they even doubled down a year later with 14th gen pushing clocks yet higher again, increasing stability issues.

Well, the newly manufactured Raptor Lake has no problems...
In the first place, there are SKUs with Problems and SKUs without Problems
It's not that Raptor Lake has a problem as a whole...

There was no problem with the architecture...
Well, there is a problem that it consumes a lot of power.

 

[Xebec]

Well, the newly manufactured Raptor Lake has no problems...
In the first place, there are SKUs with Problems and SKUs without Problems
It's not that Raptor Lake has a problem as a whole...

There was no problem with the architecture...
Well, there is a problem that it consumes a lot of power.

13th and 14th Gen Intel CPU instability also hits servers — W680 boards with Core i9 K-series chips are crashing

These non-Z series motherboards don't push 13th and 14th Gen chips nearly as hard, so what's happening?

www.tomshardware.com

Unfortunately for Intel, the stability issue also hit a lot of server based products -- causing entire server farms to just dump Intel for AMD. The brand reputation was significantly damaged by pushing clocks and voltage margins too hard. The root cause appears to have been higher than expected voltage spikes from what I read - which could be a combo architecture and process issue. (i.e. my theory - process not supporting the physical needs of the architecture which is drawing power beyond what predicted).