Discussion Intel - the cost of BACKPORTING

Kocicak

Senior member
Jan 17, 2019
982
973
136
Seeing the preliminary results in the Anandtech 11700K review, the performance of this CPU compared to the previous CPU overall seems to be just marginally better, in some cases even worse than the previous generation product.

What is the economic sense and impact on the company of the decision to backport this new CPU to the 14 nm technology? How much money does this backport actually cost? Was this cost really worth it, when the only result is that you will have some "fill in product" for a few months before the Alder lake comes?

If it does not make sense financialy, then why did they do it? They must have known well that the end result of this will be.
 
  • Like
Reactions: krumme

jpiniero

Lifer
Oct 1, 2010
14,487
5,155
136
It's a lot of money. Like we are talking tens of millions. My opinion is that they wanted something that wasn't Skylake to sell for awhile on 14 nm. That it ended up being a fill in product is more because of the end result.

Remember they also did Rocket Lake U which did get cancelled.
 

Kocicak

Senior member
Jan 17, 2019
982
973
136
Yes, but they knew in advance what the end result will be, they are not stupid, they can simulate all this stuff.

What they did seems completely irrational to me. What would be rational in my opinion is to keep selling 10th gen products, and price them so low that they make sense perf. per dollar wise compared to AMD products.
 

TheELF

Diamond Member
Dec 22, 2012
3,967
720
126
when the only result is that you will have some "fill in product" for a few months before the Alder lake comes?
Let's not forget that this arch was supposed to come out 5 years ago and on 10nm, I'm not saying alder will not come out soon but I also wouldn't bet on it that it will.

Also there are enough things were the 11700k is quite a bit faster than the previous model and that's all still stock settings.
If it does not make sense financialy, then why did they do it? They must have known well that the end result of this will be.
It makes the most financial sense ever, keep making money on 14nm while everybody else is being forced to move to 7 and 5nm, I don't know if people realize this but there is a limit to how small you can get and that's when you can't make any money anymore.
Everybody has a terrible time at getting enough small nodes for their production while intel is raking in the money.
What they did seems completely irrational to me. What would be rational in my opinion is to keep selling 10th gen products, and price them so low that they make sense perf. per dollar wise compared to AMD products.
How is making less money making sense to you? They don't need to compete on perf. per dollar so why should they?
 

Kocicak

Senior member
Jan 17, 2019
982
973
136
It makes the most financial sense ever, keep making money on 14nm while everybody else is being forced to move to 7 and 5nm, ....
Yes, but preferably not by developing and backporting a product to it with associated (high) cost, which will have a lifespan of a few months, and moreover it will be in difficult position in terms of competing products and own products of previous generation.

Using 14nm is probably very cheap for Intel, however it makes products which are today not very competitive in performance per watt.
 
  • Like
Reactions: BadThad

Jimzz

Diamond Member
Oct 23, 2012
4,399
190
106
The biggest thing I can think of is it gives Intel time top work out any bugs from this design while they work on the next 10nm++ node. Once the refined 10nm is ready they have better design and fixes to move over.

Its a streacth but this design is several years old and they put billions into it. So they have to show something for it.
 

podspi

Golden Member
Jan 11, 2011
1,965
71
91
Of course they didn't know Rocket Lake was going to turn out this bad (assuming it did, which seems likely). Yes, there are simulations but physical implementations matter too, and they're ... physical. My guess is the physical implementation of this killed it - maybe due to the latency issues. Remember this design was intended for 10nm, not 14nm - and the original release of the core did not clock particularly well.

It isn't that hard to believe that in backporting it to 14nm and tuning for high clock speeds, something broke that couldn't be fixed in time. As it is, if Alder Lake really is coming at the end of the year this launch almost doesn't seem worth it. Can't imagine upgrading from 10th gen to this unless you had a very limited set of workloads.
 

dmens

Platinum Member
Mar 18, 2005
2,271
917
136
The sole purpose was to get features out there: i.e. AVX-512 (if you can call that a feature), and upgraded I/O standards. This is so management can play CYA with shareholders by claiming technical progress (since benchmarks don't matter), and to give the benchmarketers something to bloviate about.
 

Hulk

Diamond Member
Oct 9, 1999
4,187
1,970
136
The Rocket Lake backport is also an interesting economic decision when you consider H series Tiger Lake with 8 cores and up to 5GHz single core frequency is right around the corner. Assuming Intel currently has a limited amount of 10SF production capacity they probably would have had to transition a 14++++++++++++++++ fab to 10SF and perhaps that would have been even more time consuming and costly than pushing Rocket Lake out the door.

I'm thinking that this comes down to production capacity. Intel sells a lot of CPU's. They know how many orders they have to fulfill and the only way that was going to happen was by staying with the old reliable 14++++++++++++. It's one thing to fall behind with architecture. It's another thing to fall behind with process. But if you also fall behind on production then that's three strikes and you're out.
 

Kocicak

Senior member
Jan 17, 2019
982
973
136
Of course they didn't know Rocket Lake was going to turn out this bad (assuming it did, which seems likely). Yes, there are simulations but physical implementations matter too, and they're ... physical. My guess is the physical implementation of this killed it - maybe due to the latency issues. Remember this design was intended for 10nm, not 14nm - and the original release of the core did not clock particularly well.
I am not sure what you talk about when you say "physical implementation". 14 nm is possibly 10 years old process already (including research, development, use for production and all the improvements). This can be the most well known, researched and analyzed process known to mankind. Intel knows EVERYTHING about it.

CPU is such so extremelly complex system, that the simulations must run 100% well, if they did not, nobody could build a functioning CPU.

Intel knew EXACTLY what is the end result going to be.
 

inf64

Diamond Member
Mar 11, 2011
3,680
3,943
136
I am not sure what you talk about when you say "physical implementation". 14 nm is possibly 10 years old process already (including research, development, use for production and all the improvements). This can be the most well known, researched and analyzed process known to mankind. Intel knows EVERYTHING about it.

CPU is such so extremelly complex system, that the simulations must run 100% well, if they did not, nobody could build a functioning CPU.

Intel knew EXACTLY what is the end result going to be.
The bad news for intel is that they know, with great degree of certainty, what is coming from AMD's camp. That is why you see this incredible push to launch AL in Q4, almost at all cost. They know that AMD has 6nm part that will basically equal AL in gaming (more or less), have a smidge less IPC and clock higher. They also know that Zen4 will reset them back to Comet Lake Vs Zen3 scenario and they maybe have 4-6 months of sort of performance parity ( Q4 2021-Q2 2022). After that it's back to the meat grinder again, on all fronts. AMD's supply limitation is the only saving grace.
 
  • Like
Reactions: Tlh97

Spartak

Senior member
Jul 4, 2015
353
266
136
More like 18 months ago.

No way in hell it was just 18 months ago. Remember Rocket Lake is delayed by half a year. Add a year on top of that since you need about two years if all goes according to plan.

Not bothered to look it up but I seem to remember JIm Keller stated they succesfully decoupled process from design about half a year before the introduction of sunny/willow/golden cove in december 2018. The need for a backport was already very clear and broadly discussed here (mostly me granted but also others).

The design would have probably already started in may/june 2018 but the project greenlit would be some months later. I'm thinking october 2018 myself since that's when Charlie Demerjian reported that 10nm was cancelled, which I believe was indeed semi accurate as it related to the desktop product only. October 2018 was likely the moment through testing/simulation they felt secure enough to pull the trigger on Rocket Lake. You dont cancel a product without greenlighting a proven viable alternative.
 
Last edited:

Kocicak

Senior member
Jan 17, 2019
982
973
136
Well, the same thing about the long time delay could be said about Alder lake too.

If the Rocket was ready for 10 nm, how long could the process of backporting it to 14 nm last? I bet you do not count that time in weeks. Alder lake might have been contrarily sped up, thus shortening the life of RL from the other side.
 

dmens

Platinum Member
Mar 18, 2005
2,271
917
136
No way in hell it was just 18 months ago. Remember Rocket Lake is delayed by half a year. Add a year on top of that since you need about two years if all goes according to plan.

Not bothered to look it up but I seem to remember JIm Keller stated they succesfully decoupled process from design about half a year before the introduction of sunny/willow/golden cove in december 2018. The need for a backport was already very clear and broadly discussed here (mostly me granted but also others).

The design would have probably already started in may/june 2018 but the project greenlit would be some months later. I'm thinking october 2018 myself since that's when Charlie Demerjian reported that 10nm was cancelled, which I believe was indeed semi accurate as it related to the desktop product only. October 2018 was likely the moment through testing/simulation they felt secure enough to pull the trigger on Rocket Lake. You dont cancel a product without greenlighting a proven viable alternative.

Maybe 2 years then. Hard to nail down exactly since Intel management is so weak and dithering. Moreover, as far as I can tell, senior management was being lied to about 10nm process health. They ought to have approved this backport in 2016 when they couldn't make enough healthy Cannonlakes to paper launch a Chinese laptop, at least that would have given them to tune down the design so the power draw isn't so outrageous.

My point is that Rocketlake is the biggest rush hack job in company history, based on conversations with friends.

Also Jim Keller was referring to the decoupling of design from Intel process, and the use of external foundries. Moving a design database between internal Intel processes has been done for decades, i.e. with tick shrinks. Rocketlake is no different from that... except they moved backwards this time LOL.
 

TheELF

Diamond Member
Dec 22, 2012
3,967
720
126
Yes, but preferably not by developing and backporting a product to it with associated (high) cost,
We can't know that but if it's still cheaper than producing the same arch on 10nm then it's win win.
which will have a lifespan of a few months
We can't know that, alder will be a completely different approach to CPUs so even if it releases at the end of the year and that's a huge if, intel might still keep the traditional CPUs alongside those for years to come.
and moreover it will be in difficult position in terms of competing products and own products of previous generation.
The previous 4 generations of intel CPUs have been in a difficult position compared to the competition and they have all sold like crazy and this gen has actual IPC improvement.
Using 14nm is probably very cheap for Intel, however it makes products which are today not very competitive in performance per watt.
Again looking at the last 5 years it doesn't seem that anybody cares about that since intel's CPUs are selling like crazy.
 

Jimzz

Diamond Member
Oct 23, 2012
4,399
190
106
Again looking at the last 5 years it doesn't seem that anybody cares about that since intel's CPUs are selling like crazy.


I think a lot of that is legacy orders. Kinda like the old, but forgotten, saying of "No ever got fired for buying IBM...". That plus AMD can't make enough chips to keep up.
If AMD can get prodtion up and even a few ARM chips, like apple, take more space it will affect intel badly.

Intel is in the Cash Cow years if they do not change. They will make good money on what is a legacy based product but long term will lose out in the future.
 
  • Like
Reactions: Tlh97 and krumme

Spartak

Senior member
Jul 4, 2015
353
266
136
Maybe 2 years then. Hard to nail down exactly since Intel management is so weak and dithering. Moreover, as far as I can tell, senior management was being lied to about 10nm process health. They ought to have approved this backport in 2016 when they couldn't make enough healthy Cannonlakes to paper launch a Chinese laptop, at least that would have given them to tune down the design so the power draw isn't so outrageous.

My point is that Rocketlake is the biggest rush hack job in company history, based on conversations with friends.

Also Jim Keller was referring to the decoupling of design from Intel process, and the use of external foundries. Moving a design database between internal Intel processes has been done for decades, i.e. with tick shrinks. Rocketlake is no different from that... except they moved backwards this time LOL.


Decoupling was EXACTLY done so you could backport. Bringing it to a next gen process is much easier. It was very clear from the words used in the interview that they wanted to prevent a future repetition of stagnation in architecture when f.i. 7nm also turns out problematic.

And regarding the hard to nail down part, I think I made a pretty convincing case for the oktober 2018 time frame. Since I also turned out right about the 14nm Sunny *cough* Cypress Cove backport as the sole person around these forums back then I'll stick to my interpretation of that interview.