Zen hasn't taped out yet

[LTC8K6]

Most likely the writer simply didn't spell "taping" correctly.

 

[Techhog]

Anyone else here find the language in the transcript a bit confusing?


As an example:

"Hans Mosesmann - Raymond James & Associates.Okay. Going more into the graphics side of the business if I will talk about virtual reality, okay to say new Skylake and your profits are kind of creating an environment with visit there is an upgrade type of potentially with on the GPU side, what specifically are you guys doing to kind of benefit from that, that your…"

Go home, whoever typed this transcript. You're drunk. In fact, you're seconds away from alcohol poisoning death...

 

[Techhog]

Most likely the writer simply didn't spell "taping" correctly.

Can't we try to stay positive? ^__^;

 

[DrMrLordX]

Dude, read the transcript. First CPUs based on Zen will be HEDT (Summit Ridge) then server CPUs coming later.

Duh. APUs aren't until 2017.

My point still stands.

 

[krumme]

I think people are confused because they treat "tape-out" as a single event on don't realize it is a process of multiple tape-outs.

1. Initial Prototype Tapeouts (PLURAL)
- Once achieved, prototype silicon is usually three months away
- Once prototype silicon is on hand, testing informs modifications
- Almost always not using the full chip design

2. Engineering Sampling Tapeout(s)
- Usually five or six months after the prototype tapeouts begin
- Usually always the first full-chip design for validations

3. Production Revision Tapeout(s)
- Potentially the final steppings, addresses issues found in ES silicon
- This usually takes three months (now 9 months total)

4. Production Validation Tapeout
- Should be the final stepping, usually two or three months after step 3
- We are around the one year mark now.

5. Initial Production
- Provided #4 works smoothly production begins
- This is generally a month after #4, sometimes two

I think we are between Step 1 and Step 2, which means we have 9 to 11 months to go, barring any unforeseen occurrences.

What makes you say its between step 1 and 2?

 

[looncraz]

What makes you say its between step 1 and 2?

A combination of factors, really. The official GloFo and AMD announcements that they have silicon success for FinFet means they have accomplished #1. The timing of this is perfect, to within days, of when an announcement should have been expected if the initial rumors of Zen's tape-out were correct and the rumors of Zen going into prototype production were correct, the timing there also being absolutely perfect.

When things line up like that in the rumor mill and the data is coming from different sources, and one of them dead-set official, it strongly suggests that Zen is what has been prototyped. Not to mention just how important Zen is for AMD, and Su's previous declarations that Zen was their priority.

From Global Foundries' own mouth:

"AMD has taped out multiple products using GLOBALFOUNDRIES’ 14nm Low Power Plus (14LPP) process technology and is currently conducting validation work on 14LPP production samples."

AMD just doesn't have many bleeding-edge products going, so it's pretty safe to assume that Zen, in some form. is one of those products.

 

[Systems analyst]

I think people are confused because they treat "tape-out" as a single event on don't realize it is a process of multiple tape-outs.

1. Initial Prototype Tapeouts (PLURAL)
- Once achieved, prototype silicon is usually three months away
- Once prototype silicon is on hand, testing informs modifications
- Almost always not using the full chip design

2. Engineering Sampling Tapeout(s)
- Usually five or six months after the prototype tapeouts begin
- Usually always the first full-chip design for validations

3. Production Revision Tapeout(s)
- Potentially the final steppings, addresses issues found in ES silicon
- This usually takes three months (now 9 months total)

4. Production Validation Tapeout
- Should be the final stepping, usually two or three months after step 3
- We are around the one year mark now.

5. Initial Production
- Provided #4 works smoothly production begins
- This is generally a month after #4, sometimes two

I think we are between Step 1 and Step 2, which means we have 9 to 11 months to go, barring any unforeseen occurrences.

This looks correct to me. It is consistent with what was said by AMD:

1. [Initial] Tape-out occurred a few months ago. This is stated explicitly.
2. Development is underway. Stated explicitly.
3. Samples will be available to developers in a few months, for validation. Stated explicitly.
4. Production chips are usually available 12-14 months after [initial] tape-out. Stated explicitly.
5. Production will ramp thereafter. Stated explicitly.

This is consistent with availability of some production chips in 2016Q4.

It seems to me that some people in this thread have an unreasoning dislike of AMD; they wilfully misread AMD's statements, to support their own private view of reality. Personally, I am neutral on the subject.

Only time will tell whether AMD deliver on this slightly vague schedule.

 

[LTC8K6]

A combination of factors, really. The official GloFo and AMD announcements that they have silicon success for FinFet means they have accomplished #1. The timing of this is perfect, to within days, of when an announcement should have been expected if the initial rumors of Zen's tape-out were correct and the rumors of Zen going into prototype production were correct, the timing there also being absolutely perfect.

When things line up like that in the rumor mill and the data is coming from different sources, and one of them dead-set official, it strongly suggests that Zen is what has been prototyped. Not to mention just how important Zen is for AMD, and Su's previous declarations that Zen was their priority.

From Global Foundries' own mouth:

"AMD has taped out multiple products using GLOBALFOUNDRIES’ 14nm Low Power Plus (14LPP) process technology and is currently conducting validation work on 14LPP production samples."

AMD just doesn't have many bleeding-edge products going, so it's pretty safe to assume that Zen, in some form. is one of those products.

Hmmm...

GLOBALFOUNDRIES today announced it has demonstrated silicon success on the first AMD (NASDAQ: AMD) products using GLOBALFOUNDRIES’ most advanced 14nm FinFET process technology.

Today’s announcement represents another significant milestone towards reaching full production readiness of GLOBALFOUNDRIES’ 14LPP process technology, which will reach high-volume production in 2016.

“FinFET technology is expected to play a critical foundational role across multiple AMD product lines, starting in 2016,”

What new AMD chips would be in "high-volume production" in 2016?

It doesn't seem like that would be Zen.

Zen seems like high volume in 2017.

 

[maddie]

Hmmm...




What new AMD chips would be in "high-volume production" in 2016?

It doesn't seem like that would be Zen.

Zen seems like high volume in 2017.

I'm pretty ignorant of the present state of upcoming AMD products, but is it confirmed that Bristol Ridge/Carrizo on Am4 will be 28nm.

 

[jpiniero]

Hmmm...
What new AMD chips would be in "high-volume production" in 2016?

The consoles. Didn't I mention this earlier in this thread?

 

[Andrei.]

You are also talking about: it runs our 1 version of the OS and a couple built in applications without any shockingly visible errors so ship it.

Real CPU validation is a bit more in depth with things like: does it run all legacy OSes without error, does it run all versions of that legacy OSes without error, does it do that across different OSes, and across different applications (LOTS AND LOTS of applications). Then does it work with all these hardware addons, etc.

single purpose socs might get released in 4 months after silicon, but it is only because they've had a fraction of the validation, and in my experience, it certainly shows up in device stability.

Thanks for the explanation, but there's plenty more validation in mobile too. There are well well other mobile systems (Windows), plethoras of hardware interfaces and most importantly a whole cell modem that needs to work as designed. You think that all doesn't need complex validation?

I don't see Intel CPUs having any less erratas than ARM ones.

As for how this industry works, my name is on numerous patents and I was one of the lead architects for QPI, so I think I've got just a bit more of a clue.

Such posts are unwarranted. I'm still waiting on an explanation/reply from you in regards to this post where you're most definitely wrong.

 

[ElFenix]

Today’s announcement represents another significant milestone towards reaching full production readiness of GLOBALFOUNDRIES’ 14LPP process technology, which will reach high-volume production in 2016.

What new AMD chips would be in "high-volume production" in 2016?

that quote you picked does not say that AMD products will be in high volume production on GF 14LPP. it says that GF 14LPP will be in high volume production.



edit: i don't know why we have 2 current zen threads.

 

[pm]

Thanks for the explanation, but there's plenty more validation in mobile too. There are well well other mobile systems (Windows), plethoras of hardware interfaces and most importantly a whole cell modem that needs to work as designed. You think that all doesn't need complex validation?

I don't see Intel CPUs having any less erratas than ARM ones.

These 12-18 schedules for tape-out to public release match with my experience as well, and I've worked at 3 CPU companies (Inmos, HP and Intel). In fact, in high-end server products, I've seen substantially longer than even these numbers. I've also observed industry data which agree wtih the much shorter times for mobile SOC's that you are quoting, Andrei, but in these the main core is delivered as a IP that's already been validated. In the shortest mobile SOC post-tape-in validation cycles they are getting the core and substantial portions of the design delivered as a completely hardened IP (ie. GDS layout).

In the mobile space, it appears to me that ARM is shouldering the bulk of the electrical and functional validation effort prior to the tape-in of the external SOC product. So I think you have to add this core validation to the front of any mobile schedule that you are looking at and then the respective numbers start to make more sense. Alternatively, you could picture that the big old stodgy companies aren't as nimble at developing products as the new small ARM-core-using mobile SOC companies... but I tend to think it's more of the former... that a lot of the heavy-lifting for validation is done by ARM or it's early launch partners.

Patrick Mahoney
Senior Design Engineer
Intel Corp.
* Not an Intel spokesperson... just an engineer who happens to work there *

 

[Andrei.]

Andrei, but in these the main core is delivered as a IP that's already been validated.

That's what I suspected - thank you very much for the confirmation on the matter. Ultimately we have no idea how long ARM's validation process takes, they've told me a CPU IP is basically in development for about 4-5 years. The thing about that strikes me as odd is that sometimes they don't have internal silicon - meaning it's also a very tightly nit collaboration with the lead partners of a given CPU arch.

electrical and functional validation effort prior to the tape-in

You say prior to tape-in. But what about validation past tapeout? The discussion came about because AMD is supposedly taking long to validate post-tapeout.

 

[pm]

That's what I suspected - thank you very much for the confirmation on the matter. Ultimately we have no idea how long ARM's validation process takes, they've told me a CPU IP is basically in development for about 4-5 years. The thing about that strikes me as odd is that sometimes they don't have internal silicon - meaning it's also a very tightly nit collaboration with the lead partners of a given CPU arch.

You say prior to tape-in. But what about validation past tapeout? The discussion came about because AMD is supposedly taking long to validate post-tapeout.

Yeah, my wording could have been clearer. What I was trying to say is that if you are a vendor with a mobile SOC that uses an ARM core then I believe you will get the validation feedback on that core from ARM prior to the tape-in of your SOC project so that you have time to make the changes before you commit them to the mask. While you are still designing your SOC, you are getting the bug/errata data from ARM about the problems and getting fixes for them. But it should be noted that when it comes to figuring out how ARM and it's partners design things, I'm just guessing. About the only thing that I know for a fact is that the original question you had - are these quoted post-design validation numbers realistic - that they are, in my experience, completely realistic... and maybe even shorter than they might be for a server product.

 

[mysticjbyrd]

Hmmm...




What new AMD chips would be in "high-volume production" in 2016?

It doesn't seem like that would be Zen.

Zen seems like high volume in 2017.

Wouldn't they be in high production in 2016, so they could be fully available in 2017?

 

[Dresdenboy]

At least such vague statements are good to stir up a lengthy discussion. Human nature doesn't like soft facts with attached probabilities.

Let me add a thought:
How does validation of tuned synthesized designs work? Does detailed macro simulation aid the process before having silicon?

I didn't touch that topic in Chinnery and Keutzer yet.

 

[witeken]

Patrick Mahoney
Senior Design Engineer
Intel Corp.
* Not an Intel spokesperson... just an engineer who happens to work there *

Cool. Really like insider information from engineers working there.

 

[DrMrLordX]

I'm pretty ignorant of the present state of upcoming AMD products, but is it confirmed that Bristol Ridge/Carrizo on Am4 will be 28nm.

Yeah, pretty much. It's supposed to be GF28A, not 28 shp though. So, not the same as Kaveri, but rather Carrizo.

 

[mrmt]

Yeah, that also struck me as odd. Is the transcript writer and/or CFO really not familiar with the concept "taping out" so they by mistake said "tapping out"? Or did they in fact mean "tapping out", indicating something like the tap being turned on at the factory to start producing ES chips!?

Kind of confusing all of this I have to say. A clarification from AMD would be welcome.

Kumar already made some blunders in Q&As with investors, so I wouldn't be surprised if he is wrong on this one too, especially because it would be kinda hard to reconcile his statement of Zen revenue in 2016 with the chip not taping out in 2015. One of them is patently wrong.

In any case, I think this subject will be raised in the quarterly Q&A next month and Lisa Su will be in charge, so at least the answer should be more straightforward.

 

[LTC8K6]

Wouldn't they be in high production in 2016, so they could be fully available in 2017?

If high volume is reached in December 2016, then yes I would expect availability in 2017.

If reached in November 2016 or earlier, then I would expect availability in 2016.

 

[LTC8K6]

that quote you picked does not say that AMD products will be in high volume production on GF 14LPP. it says that GF 14LPP will be in high volume production.



edit: i don't know why we have 2 current zen threads.

Then you agree with me that the GF press release doesn't necessarily indicate anything at all about Zen?

 

[ElFenix]

Then you agree with me that the GF press release doesn't necessarily indicate anything at all about Zen?

I haven't read the whole thing, so no, at this point I can't. But the answer is really 'null' rather than 'no.'

That said, I would think it would be odd if a supplier were unilaterally discussing the products of the firm it is supplying, even despite the relationship between GF and AMD.

 

[maddie]

The language in the report has so many problems that I find it difficult to place much importance to it.

Someone, please parse this question in the report for me.


"Hans Mosesmann - Raymond James & Associates.Okay. Going more into the graphics side of the business if I will talk about virtual reality, okay to say new Skylake and your profits are kind of creating an environment with visit there is an upgrade type of potentially with on the GPU side, what specifically are you guys doing to kind of benefit from that, that your…"

 

[mrmt]

The language in the report has so many problems that I find it difficult to place much importance to it.

Someone, please parse this question in the report for me.

You are free to make a new transcript and upload to Seekingalpha:

http://ir.amd.com/phoenix.zhtml?c=74093&p=irol-eventDetails&EventId=5208072