Question Should Intel have skipped Rocket Lake and going straight to Alder Lake?

[Hulk]

Okay let's play armchair quarterback. Knowing what we do now regarding Rocket Lake and Alder Lake, should Intel have skipped Rocket Lake and instead just stayed with Comet Lake to keep 14nm production going until Alder Lake was ready? All of the resources devoted to that backport could have gone to other less "stop-gappy" measures. The gains at the top of the stack moving from 10 core Comet to 8 core Rocket were quite minimal and in some cases nonexistent. It's not like Rocket was a Zen 3 beater anyway.

I'm just wondering if some of the guys at Intel who struggled over this decision years ago are looking back today and thinking "we probably should have skipped it."

 

[CakeMonster]

In hindsight... maybe. But they probably had lots of OEM commitments to fulfill that made them money, and I bet it was planned at least a couple of years before they knew exactly how/when AL would turn out. I'm not at all convinced the resources were 'wasted' in that sense.

 

[Zucker2k]

Rocket Lake had a window of relevance, unfortunately, by the time it was launched that window was almost shut. I believe the proponents of the idea were calling for it as far back as 2017, so you could've imagined the reception OG Zen would've received in benches. Having said that, I'm not entirely sure Intel would've gone straight to 8 cores, however, a 6 core Rocket Lake would've still rekt the 1800x, and Intel would've been happy to sell it for $499 back then.
Still, even in 2021, Rocket Lake has been competitive against the 4th iteration of Zen core for core, and in single/less threads scenarios, thanks to it's whopping higher single/double core frequencies.

 

[Accord99]

I think in hindsight given that Intel finally mastered 10nm with Tiger Lake, they should have used all that resource into getting the the 8-core Tiger Lake released earlier and on the desktop. But it's understandable why Intel would be gun shy with a 10nm product given the trouble they had for years.

 

[Hulk]

I think in hindsight given that Intel finally mastered 10nm with Tiger Lake, they should have used all that resource into getting the the 8-core Tiger Lake released earlier and on the desktop. But it's understandable why Intel would be gun shy with a 10nm product given the trouble they had for years.

The theory around here goes AFAIK that at that point in time Intel didn't have a lot of 10nm capacity so what 10nm they did have needed to go to mobile. Additional 10nm, namely 10ESF or now Intel 7 wouldn't be ready until Alder Lake. The architecture wasn't the problem holding things up it was the process technology.

So the choice was soldier on with Comet Lake or go with the backport to Rocket Lake. I'm just thinking that in hindsight all of that effort could have been put to better use as it was really just a matter of 7 or 8 more months with Comet Lake.

 

[NTMBK]

They should have put the effort into porting to TSMC 7nm or 5nm instead...

 

[diediealldie]

Intel had tons of 14nm capacity due to 10nm migration failures. Intel was fighting with "moar-coar" tactics to fight Zen. 4 core i7 became 6 core, then 8 core...etc. Meanwhile, due to an industry change, Intel had to experience backporting(or process-neutral designs).
Rocket lake was a somewhat desperate measure but served its purpose. Now a mobile market is flooding with 10nm Tiger Lake and Entry level Xeon is filled by Rocket lake Xeon.

After they go EOL, they'll be used for Chipsets and SoC. Then Intel's EMIB will shine then.

 

[DrMrLordX]

Rocket Lake had a window of relevance, unfortunately, by the time it was launched that window was almost shut.

Agreed, it launched too late.

They should have put the effort into porting to TSMC 7nm or 5nm instead...

TSMC probably wouldn't have given them enough capacity to handle Rocket Lake-S or a hypothetical IceLake-S. It would have cut into their wafer allocation for DG2 and Ponte Vecchio.

 

[jpiniero]

I think in hindsight given that Intel finally mastered 10nm with Tiger Lake

Not entirely. Intel said 70 mill Tiger Lake shipped this year, that sounds like a lot until you realize Intel sells close to that a quarter. Desktop is still a ton of sales at this point and it's a lot harder to hide volume as opposed to when it's BGA and you're buying the whole laptop and not just a processor.

Even now Intel sells a ton of Comet Lake H laptops and there's still some Comet Lake U. And of course all those Chromebook Atoms. You're not getting rid of 14 nm that easily.

 

[Hulk]

You're not getting rid of 14 nm that easily.

Sad but true!

 

[SAAA]

They should have ported golden cove if tiger and ice were truly ready and done arch-wise by 2018 but 10 nm wasn't. Instead launching in 2021 old tech on an ancient 14 nm node feels stupid in hindsight.
Not like alder is that much bigger than tiger given the die size estimates so... slighlty fatter Rocket golden-based would at least have rocked serious ST scores. No need for AVX512 too, or small cores if they weren't ready.

 

[Mopetar]

Honestly their 14nm could stick around for a long time and serve plenty of useful purposes. TSMC still gets most of their revenue from older nodes and there's no reason that Intel couldn't sell their highly refined 14nm node to all kinds of other companies if they can't find enough internal uses for it.

 

[dullard]

My computer was fried in a power outage this spring (at least the PSU, motherboard, and video card were destroyed--maybe other things too but I didn't try to revive them). At the time, there were almost no video cards for ready money and those that were available were sky high. But Rocket Lake was sitting there ready with integrated graphics. So, I ended up with a 11700 that I would never have otherwise purchased. I could have potentially gotten a 10700 instead for roughly the same price. But Rocket Lake does have a good 10% lead over Comet Lake on most CPU tasks unless you compare i9 to i9. The Rocket Lake vs Comet Lake GPU tasks are not even close. For people like me, Rocket Lake is not a bad choice. I could have done a bit better with AMD, but I would have paid nearly 50% more for the whole computer at the time due to terrible GPU prices.

I think it is wise for companies to have conditional product lines. If Alder Lake was delayed even further, Intel would have had virtually nothing to sell. Plus, Rocket Lake takes the pressure off the Intel 7 manufacturing lines. They can sell every functional chip they make on both 14 nm and Intel 7 -- so why not make both products decent?

 

[Roland00Address]

Okay let's play armchair quarterback. Knowing what we do now regarding Rocket Lake and Alder Lake, should Intel have skipped Rocket Lake and instead just stayed with Comet Lake to keep 14nm production going until Alder Lake was ready? All of the resources devoted to that backport could have gone to other less "stop-gappy" measures. The gains at the top of the stack moving from 10 core Comet to 8 core Rocket were quite minimal and in some cases nonexistent. It's not like Rocket was a Zen 3 beater anyway.

I'm just wondering if some of the guys at Intel who struggled over this decision years ago are looking back today and thinking "we probably should have skipped it."

As Ian of Anandtech, via his sideproject TechTechPotato on YouTube reminded people on youtube, if a foundry does not cross a threshold sometimes 85% use capacity then the entire process may not be profitable. Sure 14nm probably passed the profitability years ago, but you need to keep those foundries full even if you simultaneously need to move as fast as possible to the next feature geometry.

Ian Interviews #12: Pat Gelsinger, Intel CEO (Q&A Roundtable), note many of the questions are not from Ian but Ian read them off and credited since audio was less good for the questioners. From Oct 30th aka 2 days ago at the time of this post.

 

[Hulk]

As Ian of Anandtech, via his sideproject TechTechPotato on YouTube reminded people on youtube, if a foundry does not cross a threshold sometimes 85% use capacity then the entire process may not be profitable. Sure 14nm probably passed the profitability years ago, but you need to keep those foundries full even if you simultaneously need to move as fast as possible to the next feature geometry.

Ian Interviews #12: Pat Gelsinger, Intel CEO (Q&A Roundtable), note many of the questions are not from Ian but Ian read them off and credited since audio was less good for the questioners. From Oct 30th aka 2 days ago at the time of this post.

Yes I understand. My question to ponder was if Intel in hindsight should have skipped Rocket and just kept those 14nm foundries cranking out Comet parts until 10ESF was ready for the November 4 Alder Lake release? Doesn't seem like Rocket Lake was well received, actually Intel got a lot of heat from it, perhaps more than they would have received if they'd just stayed with Comet Lake until Alder. The tradeoff being not having to fund the backport and spent that money on more "forward-looking" endeavors.

 

[jpiniero]

The alternative really is they could have just updated the IGP while sticking with Skylake.

 

[Roland00Address]

Yes I understand. My question to ponder was if Intel in hindsight should have skipped Rocket and just kept those 14nm foundries cranking out Comet parts until 10ESF was ready for the November 4 Alder Lake release? Doesn't seem like Rocket Lake was well received, actually Intel got a lot of heat from it, perhaps more than they would have received if they'd just stayed with Comet Lake until Alder. The tradeoff being not having to fund the backport and spent that money on more "forward-looking" endeavors.

My bad, I skipped some steps explaining the logic and that is my fault in explaining.

——————

Ian in a different video argued (April 1st, not an April fools joke or clickbait, that ”Rocket Lake is a Success for Intel 🚀💯”)

For many of the goals was keeping engineering teams busy and better learn the techniques to backport and use multiple processes for a design to better implement the dual TSMC but also Intel fabs future, yet also different foundries geometries even if it’s a single company technology.

——

Did Rocket Lake sales by itself justify the expense? Ian says he does not know but we can argue this till the ”cows come home”, his point is Intel had multiple goals and the practice backporting and using multiple proces geometries, and even different company foundries is Intels new goal it is going to keep on doing. Judging it based off prior arguments of how Intel did business and judge success no longer applies for Intel is shifting strategies.

For several reasons which the April video does not full explain, one is the licensing its cores to 3rd parties like having a TSMC fab core with other IP attached that is not Intel, but also you may want to do something crazy like have a core on Intel 4 / 7 and possibly a third geometry simultaneously to maximize fab utilization with the same gen of core.

How Intel did business for 2 decades prior is no longer how Intel thinks it is going to do business in the future. In my own words here (not Ian’s) Intel is getting weird. An understandable form of weird but not similar to Intel’s past tradition and culture and you have to look outside of how other companies design chips and not prior Intel’s methods.

 

[IEC]

Rocket Lake isn't bad. It's just too little, too late. It's not a significant enough upgrade for most Intel users and it's not very power efficient either. The hex core parts make perfectly capable budget gaming rigs (as seen below in my 11500/6600XT gaming rig).

It doesn't hold a candle to my 5950X/6800XT rig but I can travel with it and at 1080p it is good enough: