News Intel Refutes Reports of Further Roadmap Delays

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.

bit_user

Polypheme
Ambassador
I know the Wells Fargo reported on it but is also citing the SA article for that. The SA article is what started this all. No one else was stating it.
Again, you can't know that. Others surely spotted the inconsistencies.

The main difference I see between your slides is on states shipping the other states Volume Production. Could mean nothing or could mean they will ship the product in 1H 2020 with full volume production expected in 2H 2020.
I'm not exactly an industry insider, but I think it's a distinction without a difference. I think you don't ship for production before ramping volume. If anything, the volume ramp should happen slightly in advance of when you start shipping for production.

Regardless, they're now saying production begins in 2H 2020. So, forget about what "volume ramp" means, we have public statements from multiple Intel representatives that are now saying production shipments will happen later than the slide did.

And my slides? They're from this article! Don't tell me you didn't even read past the SemiAccurate reference! If you expect anyone to take you seriously, you need to do better than that.

Above, you can see that I honestly forgot about the SA reference, because the article quickly moved past that and into the real evidence of the slippage - the public statements and the slides.
 

InvalidError

Titan
Moderator
I'm not exactly an industry insider, but I think it's a distinction without a difference. I think you don't ship for production before ramping volume. If anything, the volume ramp should happen slightly in advance of when you start shipping for production.
OEMs don't jump on the new chips overnight, they phase them in across their lineup over several months or even years in some cases, that's why the last-order date for many CPUs is three or more years after the initial launch. You have to keep in mind that enthusiasts who demand zero-day or even pre-order access to all of the latest and greatest tech are only a tiny fraction of the market, most people and a large chunk of the corporate world only needs something that works, doesn't matter too much if it is one or two generations old.

Back when Intel was more openly commenting on production volumes, it took 6-8 months from launch for Intel to transition half of its production to new-gen. I doubt it is much different today. Start production of new parts several months before launch to build up process performance and binning stats, validate the entire production and verification tool chain and if everything works fine, schedule fabs for transition as needed for expected sales of old and new parts. That last part is more easily said than done when you already can't spare any fab capacity to cover for downtime and the new chips are getting bigger due to repeat delays in process shrinks. With Intel's "copy exactly" strategy, a process fully validated in one fab can be ported to its other identical fabs pretty quickly.
 

bit_user

Polypheme
Ambassador
OEMs don't jump on the new chips overnight, they phase them in across their lineup over several months or even years in some cases, that's why the last-order date for many CPUs is three or more years after the initial launch. You have to keep in mind that enthusiasts who demand zero-day or even pre-order access to all of the latest and greatest tech are only a tiny fraction of the market, most people and a large chunk of the corporate world only needs something that works, doesn't matter too much if it is one or two generations old.
We're talking about their server chips, which have a fairly stable demand and predictable service life (in the cloud, at least). And cloud operators pretty much always require the latest generation, unless it turns out to be an unmitigated disaster (or, perhaps a disaster of mitigations). The reason being that their volume of customers and the customers' load is always increasing, so if they can't continually source more efficient products, it blows holes in business models.

Anyway, it's a moot point, because only that one slide talked about volume ramps, whereas the first slide and the public statements talked about production shipments. So there's your inconsistency, without have to to pay any heed to what "volume ramp" actually means.
 

InvalidError

Titan
Moderator
We're talking about their server chips, which have a fairly stable demand and predictable service life (in the cloud, at least).
Even in the datacenter space it is quite typical to phase in new servers instead of migrating everything at once since new CPUs and platforms have to be dialed in and burn-in-tested to make sure they are up to production standards before committing to larger orders some months later.
 
Again, you can't know that. Others surely spotted the inconsistencies.


I'm not exactly an industry insider, but I think it's a distinction without a difference. I think you don't ship for production before ramping volume. If anything, the volume ramp should happen slightly in advance of when you start shipping for production.

Regardless, they're now saying production begins in 2H 2020. So, forget about what "volume ramp" means, we have public statements from multiple Intel representatives that are now saying production shipments will happen later than the slide did.

And my slides? They're from this article! Don't tell me you didn't even read past the SemiAccurate reference! If you expect anyone to take you seriously, you need to do better than that.

Above, you can see that I honestly forgot about the SA reference, because the article quickly moved past that and into the real evidence of the slippage - the public statements and the slides.

The Wells Fargo report cites the SA article so yes I can know that, especially since the SA article hit first then the Wells Fargo report which again it says it cited the SA article along with "industry checks". The SA article is heavily the influence for everything in this. If there were multiple sources not all citing the exact same article sure there could be some truth to it and it not be a rumor. But when everything is citing the same article it tells me that its just one guys speculation being pushed out and nothing more.

The TH article is just them working with Intels response and speculating from there.

As I said it could mean nothing or it could mean what I said. Considering it takes a few months for testing before we see finalized products, Dell, HPE etc have to be able to take the products and design the chassis and cooling systems after all, I would err on the side that it means what I said, that they will be shipping the product 1H and expect full volume production in 2H.

But as you said we are not insiders so we can only go based on what Intel says and any rumors or speculation are just rumors and speculation until either the time passed or Intel/AMD/whoever has stated otherwise.
 

MasterMadBones

Distinguished
Since my name was mentioned after my last message, I think I should add that I agree with both mods. Although 10nm has been delayed numerous times, that is no indication of the current state of the process. As I said before, plenty of 10nm products have entered volume production in the last year, which shows to me that the process is fine. A single source is hardly reliable. I also agree that "shipping" and "volume production" are two different things. The most recent example is AMD's Rome, which was already shipping in H1 2019, but was only "officially" released in September after volume production had started.

But what if there is actually a delay? Intel, like many companies, lays out its roadmaps in quarters or years. That means if something happens to a product last-minute, they still have up to 3 months to fix it. Some very minor issues could take maybe two weeks to fix and roll out to manufacturing, but you would still call them delays internally.
 

InvalidError

Titan
Moderator
The most recent example is AMD's Rome, which was already shipping in H1 2019, but was only "officially" released in September after volume production had started.
IIRC, large customers like Amazon got their first Rome engineering samples in late 2018. The road from first working silicon to commercial production can be pretty long with any re-spins along the way to first production-worthy silicon adding 3-6 months a pop.
 

MasterMadBones

Distinguished
IIRC, large customers like Amazon got their first Rome engineering samples in late 2018. The road from first working silicon to commercial production can be pretty long with any re-spins along the way to first production-worthy silicon adding 3-6 months a pop.
I feel like AMD had a remarkably easy time with TSMC's 7nm. The late switch from GF to TSMC only delayed Zen 2 about 4 months or so it seems.
 

bit_user

Polypheme
Ambassador
The Wells Fargo report cites the SA article so yes I can know that,
No, you can't. Unless you have a crystal ball that lets you see into an alternate universe without them, you literally cannot know. It's called a counterfactual. They are unknowable, by definition.

Your anti-SemiAccurate proclivity is noted, but it sounds to me like it goes beyond what can be justified by reason, and I can't help you with that. As far as this conversation is concerned, I suggest you just drop it, because it's not constructive or really even relevant.

The TH article is just them working with Intels response and speculating from there.
Yes, that's what I'm talking about. That's literally like 90% of the article. Just skip the first paragraph and focus on the rest.

they will be shipping the product 1H and expect full volume production in 2H.
Intel slipped a schedule to their investors. They hate that. If there were any way they could fudge their way around having to do that, they would. If they could do what you said, then they'd still be talking about 1H, like they did in the first slide.
 
Last edited:

bit_user

Polypheme
Ambassador
I think I should add that I agree with both mods
You shouldn't think of them as mods, when they're actually participating in a discussion. As much as I've debated with these guys, I credit them for their professional detachment, in their moderation activities.

I also agree that "shipping" and "volume production" are two different things.
Sure, they are. Nobody said they were the same - just debating which one comes first.

Anyway, forget all that and focus on this slide:

wum2Y3GP9vQanoezSoNjR7-650-80.png


And this official word from an Intel representative:

"Intel remains on track for delivery of the Whitley platform starting with production of Cooper Lake in H1 2020 followed by Ice Lake production in H2 2020. We are also on track to follow Whitley with the delivery of Sapphire Rapids in 2021," an Intel spokesperson said.​


Both from the article, since it seems that some people don't read very carefully. Both are also via Intel's PR, and are therefore to be considered an official statement that was carefully vetted before being issued.

If that's not slipping by at least a quarter, then we can just stop the conversation right here. Because anyone who refuses to accept Intel's own word, when it comes to bad news about Intel, is certainly beyond the grasp of my ability to appeal to reason.
 
Last edited:

InvalidError

Titan
Moderator
Because anyone who refuses to accept Intel's own word, when it comes to bad news about Intel, is certainly beyond the grasp of my ability to appeal to reason.
If Intel had reasons to be optimistic about future process iterations, its new process/architecture roadmap wouldn't feature back-port strategy for each design in case it runs into any more process issues. Looks like Intel is gearing up for a really bumpy road ahead.
 
No, you can't. Unless you have a crystal ball that lets you see into an alternate universe without them, you literally cannot know. It's called a counterfactual. They are unknowable, by definition.

Your anti-SemiAccurate proclivity is noted, but it sounds to me like it goes beyond what can be justified by reason, and I can't help you with that. As far as this conversation is concerned, I suggest you just drop it, because it's not constructive or really even relevant.


Yes, that's what I'm talking about. That's literally like 90% of the article. Just skip the first paragraph and focus on the rest.


Intel slipped a schedule to their investors. They hate that. If there were any way they could fudge their way around having to do that, they would. If they could do what you said, then they'd still be talking about 1H, like they did in the first slide.

The Wells Fargo report itself states it cites the SA article. The Wells Fargo report itself came out after the SA article. The Wells Fargo report does not include any other sources except "industry checks" which is a vague term. There are no other sources in the report that I could find so unless you have some my best guess is that the SA article is what spurred the Wells Fargo report to even start.

And my original response was that Intel has already shipped 10nm and has stated that nothing has changed.

And I guess I could be anti-SemiAccurate only because its Charlies site and having been around long enough to see how he is and after he has released multiple wrong article I see anything he claims about as valid as the daily rumors that come out of WCCFTech.

If Intel had reasons to be optimistic about future process iterations, its new process/architecture roadmap wouldn't feature back-port strategy for each design in case it runs into any more process issues. Looks like Intel is gearing up for a really bumpy road ahead.

Back porting doesn't mean process issues though. If I remember didn't they state that tying a design to a process was one of their issues? Its why they have delayed so many design releases? I would think a design should be rolled out and then they should be able to apply it to whatever process best fits the need, right?

AMD has a design that currently works very well for adding more cores. I wonder how a monolithic 16 core design would do or if the I/O was not a distinct chip how it would do in heat and power or scaling for core count. Would it have been as easy for them to throw out a 16 core part for the mainstream. While Intel could go that route, there are still caveats to that design.
 

InvalidError

Titan
Moderator
Back porting doesn't mean process issues though. If I remember didn't they state that tying a design to a process was one of their issues? Its why they have delayed so many design releases? I would think a design should be rolled out and then they should be able to apply it to whatever process best fits the need, right?
New designs get delayed due to process because a smaller and faster process is required to offset the higher design complexity so clock frequencies can be maintained if not improved. In the tick-tock days, Intel migrated a design to a smaller process to dial-in the process without having to deal with a significantly different architecture, then upgraded the architecture once it knew exactly how much improvement it could afford to bake in on a given process.

As I have written before, back-porting more complex 10nm designs to 14nm will likely come with significant clock frequency and die area handicaps, it isn't something you'd want to do unless you had no other choice.

The most benign explanation could simply be that Intel does not intend 10nm capacity to ever become anywhere near sufficient to migrate all CPUs to it, so it needs to back-port some designs to meet demand for something that isn't Skylake re-re-re-refresh for the rest of the market unwilling to pay the 10nm premium or buy another iteration of the same design that now has a bunch of well-documented potentially exploitable side-channel flaws going back 10+ years.
 

MasterMadBones

Distinguished
You shouldn't think of them as mods, when they're actually participating in a discussion. As much as I've debated with these guys, I credit them for their professional detachment, in their moderation activities.
Of course, but this was the easiest way to mention both of them for me since that's what they have in common.

I also don't feel the need to repeat myself over the distinctions between some of the terminology that Intel uses that everyone here seems to agree upon. I even get the feeling that InvalidError and jimmysmitty are starting to repeat themselves.