[Digitimes] Kaveri delayed

[krumme]

Gf can probably make in order a7 consume like ooo a9. No other in business can do that.
The structural and cultural highligt of abu dhabi shines like north korea just on business class.

 

[pelov]

the real comparison should be GF's 32nm to TSMC's 28nm, and GF was 6 months ahead.

With horrendous yields and low capacity. There was some buzz around AMD losing the Apple contract because they couldn't get enough Llanos to supply Apple with MacBooks. It wasn't until the second half of the year that GloFo's 32nm was sufficient enough to see any products out of it.

What's the deal on 20nm/14nm-XM anyway? FD-SOI gate-last?

http://www.anandtech.com/show/4894/amd-confirms-32nm-yield-issues-at-global-foundries

AMD just announced revised revenue projections for Q3. Revenue is up compared to Q2 by 4 - 6%, but AMD had originally expected an increase of 10%. The reason for the revised projections? Llano supply is limited by apparently poor yields on Global Foundries' 32nm process. We had heard rumors to this effect for a while, but now they're officially confirmed by AMD.

The official statement is below:

The less-than-forecasted preliminary third quarter 2011 revenue results are primarily due to 32 nanometer (nm) yield, ramp and manufacturing issues at GLOBALFOUNDRIES in its Dresden, Germany factory that limited supply of "Llano". Additionally, 45nm supply was less than expected due to complexities related to the use of common tools across both technology nodes. AMD continues to work closely with its key partner GLOBALFOUNDRIES to improve 32nm yield performance in order to satisfy strong demand for AMD products.

They aren't the only ones to bump into process issues; TSMC and Intel aren't excluded either. TSMC was late with 28nm and highly capacity constrained for the first 6 months while Intel's 22nm products slipped by a couple of months and 14nm has/had some issues as well. People aren't ganging up on GloFo because they're the only ones with issues, it's just that they tend to run into problems more frequently

 

[MisterMac]

The seemingly slow-moving nature of the industry is probably the main reason why people have getting their heads around it, but it's just not possible to make the kind of financial and technological advances that are being demanded of GF in such a short space of time.

Some of you might enjoy this - http://www.youtube.com/watch?v=ed1WexCZgFs

As you can see the guys at GF are not sitting on their asses smoking cigars while counting AMD's revenue. They have 20nm silicon working and ready but they just can't start mass manufacturing of it until they actually have the fab ready. It's coming, there's no chance of GF not making 20nm, or 14nm after that. You just have to give them the time that it naturally takes to make it happen.

I don't disagree with the sentiment.

But every single product offering some kind of specs\performance must be compared to Market - not a singular channel of "WE DID IT!".

If your musings of near endless backing are true - then they will eventually become a gorilla as they build more and more.


However - as of right now for AMDs utopia dream scenario they're far behind.
And this is what will bleed them dry
Bulldozer failed on several accounts - one of them clearly being Leakage on the xtors.

If GF manages to create node\process tech rivalling Intel - the question is when?

And by then will it matter?
Will it matter for AMD?
I foresee by that time it will not matter for AMD - if AMD exists in it's current scenario at all.

Now here's the problem - if you say GF NEEDS AMD business they would be heavily interested in them surviving and helping them - they should do so more than crippling AMD to a limbo state.

They're treating AMD like Intel is - beneficial to keep alive.... at a "certain" level.
However not enough to pump them up or save them should something disastrous happen.

Being a soccer fan - i'll give you an example from the German Bundesliga.
A decade ago - Borussia Dortmund - was a club on the brink of Bankruptcy.
However they're "greatest" rivals Bayern Munich - loaned them alot of cash to avoid relegation\bankruptcy and let them rebuild they're club and now both are one of the top teams in Europe.

 

[SiliconWars]

The problem GF has is they are answerable to ATIC who are in turn answerable to Mubadala who are in turn answerable to the government of Abu Dhabi.

So the Abu Dhabi government sets out revenue targets for Mubadala, who in turn set out targets for ATIC (who contributes almost 50% of Mubadala's revenue), who then sets a target for GF (who contributes the vast majority of ATIC's revenue).

Even though the Abu Dhabi government has a near bottomless pit of cash, they expect to see results before handing out next years cash. The upshot of it is, GF is more likely to improve AMD's position by getting more investment long term, than it is from giving free wafers to AMD short term. If Mubadala goes to Abu Dhabi and shows them a $billion loss, do you think Abu Dhabi is going to put more or less money in to them? The answer is less, and they'd be told to work harder with less money. This is how they operate in that part of the world.

GF have accepted a WSA for 2013 that was down a third in value compared to 2012. That's a half-billion revenue shortfall they have accepted in 2013, just because AMD can't afford it. Where does the charity end? It's just amazing that people think GF is deliberately screwing AMD for everything they have - if that was the case why do they even accept WSA after WSA that is designed to help ease AMD's burden?

I'm as big an AMD fanboy that you'll find, but it's crystal clear to me that it is AMD who must improve.

 

[pelov]

GF have accepted a WSA for 2013 that was down a third in value compared to 2012. That's a half-billion revenue shortfall they have accepted in 2013, just because AMD can't afford it. Where does the charity end? It's just amazing that people think GF is deliberately screwing AMD for everything they have - if that was the case why do they even accept WSA after WSA that is designed to help ease AMD's burden?

Smaller consistent lumps of cash over a period of years add up to more money than a single chunk of cash over a single year.

You're stating that Mubadala/ATIC have to answer to Abu Dhabi via the red and black ink, but you're failing to tie the two strings together between AMD and GloFo. GloFo doesn't want AMD gone, but they also want as much money as possible. If modifying WSA's is what keeps AMD around then you dig deep and get it done.

GloFo isn't deliberately screwing AMD, they're just looking after their own bottom line.

 

[SiliconWars]

Smaller consistent lumps of cash over a period of years add up to more money than a single chunk of cash over a single year.

You're stating that Mubadala/ATIC have to answer to Abu Dhabi via the red and black ink, but you're failing to tie the two strings together between AMD and GloFo. GloFo doesn't want AMD gone, but they also want as much money as possible. If modifying WSA's is what keeps AMD around then you dig deep and get it done.

GloFo isn't deliberately screwing AMD, they're just looking after their own bottom line.

They should be looking after their own bottom line, they aren't in the business to prop up AMD, but they also aren't so short sighted that they think screwing AMD for their last dollar is the best policy either. They realise that AMD is in a bad position and they really do everything the can to ease their burden within the limits they have. I really don't see what else they can do? TSMC isn't handing out help to Nvidia, in fact they moved Nvidia to a pay per wafer agreement last year. Yet people say GF is screwing AMD? It makes no sense to me, none at all.

 

[pelov]

That has much to do with the longterm WSA agreement that will see AMD and GloFo joined at the hip for another decade. Coupled with the agreement, GloFo has been both late and capacity-constrained on their 45nm, 32nm, and now 28nm nodes.

The 'screwing' is going to be at the 20nm and 14nm-FinFETs nodes, imo. While GloFo's other customers will have access to both of those nodes early on, AMD has no choice but to wait it out to see if there will be an HPP available. This screwing has more to do with AMD signing a longterm agreement than it does GloFo. GloFo is just adjusting to what the market desires in bringing LP processes to market earlier (AMD would like the HPP early), but that's not GloFo's fault. If this was IBM, TSMC, Samsung, or Intel that signed that agreement with AMD, the end result would have been the same: a delayed high-power process.

IDC hit the nail on the head when he said that AMD is a fabless chipmaker without any of the benefits of being fabless. For example, if TSMC is to offer 20nm or 16nm-FinFETs w/ HPP early, AMD would have to pay out of GloFo's WSA and then pay TSMC to utilize it. A company like Qualcomm is able to avoid that double investment on two accounts: 1) they're in a booming market and make a crapload of chips, and 2) they avoid signing longterm WSAs. They can haggle with TSMC, because if TSMC doesn't want their business Qualcomm can just look elsewhere.

 

[SiliconWars]

IDC hit the nail on the head when he said that AMD is a fabless chipmaker without any of the benefits of being fabless. For example, if TSMC is to offer 20nm or 16nm-FinFETs w/ HPP early, AMD would have to pay out of GloFo's WSA and then pay TSMC to utilize it. A company like Qualcomm is able to avoid that double investment on two accounts: 1) they're in a booming market and make a crapload of chips, and 2) they avoid signing longterm WSAs. They can haggle with TSMC, because if TSMC doesn't want their business Qualcomm can just look elsewhere.

If AMD was fully reliant on TSMC they'd be completely screwed because TSMC doesn't have SOI currently and will only have one 20 nm process which is sort of "in-between", ie neither low power or high performance. We don't even know what kind of clocks TSMC would be capable of if they did have HPP anyway.

All of that really doesn't matter because TSMC will never have the spare capacity to give to AMD anyway, and certainly not the required 60K wpm on a new node. AMD would be far more likely to get 14nm HPP at Globalfoundries (I doubt that this will even exist) before TSMC could ever offer it to them.

The second part is the wrong way around - it's TSMC who has more demand than they can ever supply. Even AMD would bite off their hand for more 28nm, but it's just not available (at TSMC). Qualcomm is taking the same rates as everyone else or lumping it, either way TSMC gets paid the same because somebody else will snap up Qualcomm's allowance in a heartbeat.

As for AMD at GF, just think what 60-80k wpm would mean if only AMD had the right bloody chips to make with it. This is why it's AMD that must improve - they have the kind of capacity available to them that Nvidia and even Qualcomm or Apple would jump through hoops for.

In the end it was AMD who chose to make Bulldozer and Llano/Trinity at GF, while doing Jaguar at TSMC. I'm sure Jaguar would have been good on GF's SLP, maybe not hitting 2GHz of course but does that really matter? Intel will only be going as high at 10W with Atom, did AMD really need 25W with Jaguar?

If it's purely because of yields or time to market, AMD will still eventually be able to port Jaguar over to GF, opening up more wafers at TSMC for graphics or even just more Jaguars. There are clear benefits to AMD's relationship with GF - they just need the right bloody chips! This is the only real problem with AMD and GF's agreement - FX and APU's are in lowering demand because of performance and the declining PC market.

*IF* AMD had the proper chips for today's market they'd be in a great situation. Somehow this has been turned into GF screwing AMD for all they are worth. Very strange.

 

[pelov]

In the end it was AMD who chose to make Bulldozer and Llano/Trinity at GF, while doing Jaguar at TSMC. I'm sure Jaguar would have been good on GF's SLP, maybe not hitting 2GHz of course but does that really matter? Intel will only be going as high at 10W with Atom, did AMD really need 25W with Jaguar?

If it's purely because of yields or time to market, AMD will still eventually be able to port Jaguar over to GF, opening up more wafers at TSMC for graphics or even just more Jaguars. There are clear benefits to AMD's relationship with GF - they just need the right bloody chips! This is the only real problem with AMD and GF's agreement - FX and APU's are in lowering demand because of performance and the declining PC market.

AMD had the original Temash/Kabini on GloFo's 28nm (HP-m?) but it was late so they went to TSMC. AMD also had Llano on time but GloFo was late with a viable 32nm process. Moving Jaguar-based chips back to GloFo would be like shooting yourself in the foot, and then shooting yourself in the other foot

 

[SiliconWars]

Well I dunno how "late" 32nm actually was tbh. June 2011 was 2 1/2 years after Shanghai/Phenom launch. Intel for example was 2 years 4 months between Sandy (actually Westmere) and Ivy Bridge. For some reason that extra 2 months is treated like a huge delay when the reality is it wasn't really that bad. Ok the process itself was pretty whacked but again it wasn't GF's decision to go gate first, and there were plenty of rumours that it was the graphics that was the issue anyway.

But lets lay the blame on GF for Llano anyway...what about Trinity? Was that GF's fault too? The 5800K compares pretty evenly vs 980 BE with two fairly major exceptions. First is it's a 125W Phenom vs 100W Trinity, and second is Trinity has the large graphic core. It should be obvious by this stage that GF's 32nm was in pretty decent shape, but that didn't save AMD when the market fell apart at the end of last year.

Nope, I'm afraid AMD is to blame here for the disaster that was Bulldozer.

So moving back on topic, and Kaveri is delayed again. We know GF has 28nm SLP, which was always going to be first to market. The question is, is it a delay with HPP or is it a delay at AMD? I find it hard to believe they'd be over 12 months later with HPP - it's not *that* much different that they'd run into a years delay - so I'm blaming AMD for this one as well.

So now, why would they move to 28nm if they were so concerned about GF's ability to make it viable soon enough? Why not just stick on 32nm SOI, something they know well enough, know is yielding well, performance is good? Did they really risk Steamroller's performance, yields and time to market on a 10% smaller die?

 

[Arkaign]

Intel for example was 2 years 4 months between Sandy and Ivy Bridge.

Are you sure you're remembering correctly?

Sandy Bridge was released January 2011.
Ivy Bridge was released April 2012.

32nm of course predated SB, is that what you meant? 32nm to 22nm was one thing, SB to IB is another.

 

[SiliconWars]

Woops yes my bad, should have been...yes Clarkdale to IB. That was 2 years 4 months I think.

 

[Arkaign]

Woops yes my bad, should have been...yes Clarkdale to IB. That was 2 years 4 months I think.

True true. IIRC, weren't there some additional tweaks to SB's 32nm? Still 32nm of course, but I think they messed with it a bit.

 

[SiliconWars]

The big difference was the graphics core moving on-die at 32nm, Clarkdale's was packaged 45nm graphics.

 

[pablo87]

I guess Mubadala is some sort of political controlled looking by the crazy behavior and the fantacy ppt gf provides for the emperor.
If thats the case differentiating between revenue and profit can sometimes be a challenge. GF generate a lot og big numbers.
But at some time the political focus changes quicly.
I dont see the political interests in GF now:
The ambition to build a fab the worst place on earth outside of the poles have ended. What gf provides now is just big numbers until a new playing field is found.
I can see better use of the money from a political perspective. I think its a disaster in all dimensions.

Let's hope so. A leak of their accumulated losses would probably do it though nobody cares enough probably.

You may recall that AMD changed their accounting for their GF investment to original investment so that they wouldn't indirectly be disclosing GF's losses (and eventually the share certificate was just handed over of course). They boast of revenue growth which is fine but some of that is WSA contractual revenue they didn't even have to produce anything for.

Kaveri as an APU makes no sense, its just there to ensure Gloflo actually does something for their $1B+ annual cheque.

 

[mrmt]

The 'screwing' is going to be at the 20nm and 14nm-FinFETs nodes, imo. While GloFo's other customers will have access to both of those nodes early on, AMD has no choice but to wait it out to see if there will be an HPP available.

To give credit to AMD management, what they are proposing as a vision, a smaller, leaner, lagging edge, low price high volume embedded business company, is a perfect match for that subpar foundry partner of them.

 

[pablo87]

They haven't done it because they don't have to. Intel's development hasn't just outstripped the competition, it's outstripped demand.

In the past, that lack of urgency led to Intel getting overtaken. Time will tell if that will happen again.

The consumer PC business is down 20%, and there are no more significant emerging markets remaining, are you saying they should wait until the business its gone altogether?

 

[pablo87]

Isn't the GPU lane for kabini PCIe 2.0 x4?

Fine for ultra low power/ low power devices but for desktop more lanes would be nice.

I don't know the details but is it really rocket science to go from x4 to x8 for example (thus doubling the bandwidth)?

 

[SiliconWars]

Kaveri as an APU makes no sense, its just there to ensure Gloflo actually does something for their $1B+ annual cheque.

How do you propose AMD makes up the shortfall without their desktop and higher ASP mobile share?

 

[mrmt]

You may recall that AMD changed their accounting for their GF investment to original investment so that they wouldn't indirectly be disclosing GF's losses (and eventually the share certificate was just handed over of course). They boast of revenue growth which is fine but some of that is WSA contractual revenue they didn't even have to produce anything for.

It has nothing to do with disclosing GLF losses. As the IP share agreement with Intel did not allow AMD to manufacture its chips in external factories, AMD had to engineer the transaction trying to disguise Mubadala ownership. In that case, as Globalfoundries was AMD subsidiary it has to consolidate the financial statements.

After the settlement, when Globalfoundries was officially recognized as an independent company, this requirement dropped and because of that AMD did not have to recognize GLF results on its balance sheet but instead as an investment.

 

[TerryMathews]

The consumer PC business is down 20%, and there are no more significant emerging markets remaining, are you saying they should wait until the business its gone altogether?

No I'm saying they are going to wait until market forces (their competitors) compel them to offer more cores at a lower price.

The low end of the spectrum is all about what is the slowest cheapest thing we can throw together that the OEMs will buy.

Keep in mind too, the OEMs want this. They want segmented product lines because then they can upsell at a profit.

 

[pablo87]

It has nothing to do with disclosing GLF losses. As the IP share agreement with Intel did not allow AMD to manufacture its chips in external factories, AMD had to engineer the transaction trying to disguise Mubadala ownership. In that case, as Globalfoundries was AMD subsidiary it has to consolidate the financial statements.

After the settlement, when Globalfoundries was officially recognized as an independent company, this requirement dropped and because of that AMD did not have to recognize GLF results on its balance sheet but instead as an investment.

I understand this but I was referring to the subsequent change in accounting policy - I refer you to AMD's balance sheet in which their share of the GF investment was valued at $270MM in Q1 2010, $186MM in Q2 2010, $0 in Q3 2010 and then reverted back to the original $484MM "investment" in Q1 2011 when they changed accounting policy. I believe they had to otherwise they would have had to explain how their share of GF went from $484MM to $0 in 9 months (they had already stated a couple of times that they didn't participate in Global Foundries' cash calls to shareholders).

 

[pablo87]

No I'm saying they are going to wait until market forces (their competitors) compel them to offer more cores at a lower price.

The low end of the spectrum is all about what is the slowest cheapest thing we can throw together that the OEMs will buy.

Keep in mind too, the OEMs want this. They want segmented product lines because then they can upsell at a profit.

This is a plausible but it doesn't change the fact that this approach no longer works and hasn't worked for quite a while.

 

[pablo87]

How do you propose AMD makes up the shortfall without their desktop and higher ASP mobile share?

I said they needed to kill Kaveri as an APU and focus on GPU less Kaveri/Big Core CPU instead. For sub $200, they could give customers a GPU less Kaveri, mb and discrete graphics (7850) for the cost of an Intel quad core.

What is AMD's higher ASP mobile share right now, 10%? I have not been able to find a single AMD sku with discrete graphics (I'm sure there is but I tried and gave up).

 

[TerryMathews]

This is a plausible but it doesn't change the fact that this approach no longer works and hasn't worked for quite a while.

Doesn't work by what measure?

And what market do you think will improve by Intel offering cheap quad cores?

ETA: To be clear, if you're point is that Intel is anti-consumer these I will hardly disagree. If you're point is that Intel is stagnating the PC market due to a lack of effective competition, again I will not disagree. That said, Intel isn't "missing the boat" by not offering a Pentium-class quad. And if Richland picks up enough steam, I'm sure Intel will start offering Pentium-class quads.

(I believe) AMD is going to energize the low-to-mid end of the spectrum.