Richland & Kabini rumours

[Abwx]

I'm quite sure that Nvidia isn't paying TSMC $4bn/year. $2 bn maybe

edit - can't even be that high. $1bn?

Nvidia is using the most expensive process and their GPUS are not
exactly tiny pieces of silicon.

But even if it s only 1bn , it s still four times what AMD is providing ,
for sure.

Besides , slowing down AMD imply no losses on the mid term
since it will be forcibly balanced by Nvidia bigger stake on the
limited supply of the 28nm waffers.

Isnt it curious that AMD was unable to supply the demand
with the radeon 7XXX , but still , once Nvidia released their GTX6XX ,
they had no supply issue with TSMC , yet , since a waffer takes two months to be processed they should have experimented the same issues,
unless all was carefuly negotiated a few months before behind closed doors..

 

[SiliconWars]

Nvidia is using the most expensive process and their GPUS are not
exactly tiny pieces of silicon.

But even if it s only 1bn , it s still four times what AMD is providing ,
for sure.

Besides , slowing down AMD imply no losses on the mid term
since it will be forcibly balanced by Nvidia bigger stake on the
limited supply of the 28nm waffers.

Isnt it curious that AMD was unable to supply the demand
with the radeon 7XXX , but still , once Nvidia released their GTX6XX ,
they had no supply issue with TSMC , yet , since a waffer takes two months to be processed they should have experimented the same issues,
unless all was carefuly negotiated a few months before behind closed doors..

I know that TSMC is a biased fab but they don't hate on AMD. GF's behaviour towards AMD has been awful - even if it's all been above board they must realise that they are just harming themselves in the long run by holding AMD to ransom over agreements.

 

[Abwx]

I know that TSMC is a biased fab but they don't hate on AMD. GF's behaviour towards AMD has been awful - even if it's all been above board they must realise that they are just harming themselves in the long run by holding AMD to ransom over agreements.

GF could have been late for 32nm but overall AMD can only blame itself
for the current situation since they didnt learn from their 65nm debacle
that was to implement a new design on a non mastered new node.

Given Bulldozer huge complexity it was a too risky move , it would have
been better to start with improved phenom/athlon4 , moreover since
these two CPUs were still 25% of their revenues in H1 12 but alas ,
they prefered to spare a handfull of bucks not processing the masks
and keeping on wasting waffers area on 45nm while the very same
waffers could have been used for 32nm , cutting the dies area by two
and reducing the manufacturing prices to peanuts.

Another cause was that they didnt manage to advertise their
superior in die GPUs , the perception of the public to this day
is that what matters is the CPU as aknowledged by Intel s awfull
GPUs that are not even really DX11 capable but are enough for
the forcibly gullible consumer.

 

[Abwx]

It's not a straw man. It's a conspiracy theory by definition. Deal with it.

If Apple does jump ship to TSMC, which I would say is a done deal given the events of this past year, then your point is irrelevant. TSMC has bigger fish to fry.

And as far as the GloFo vs. TSMC discussion goes... Jaguar is designed to be portable. There's a strong chance you'll see it on both, dependent on how sucessful the product is.

Money talks , if Apple become the biggest customer then be assured
that they will be treated way better than Nvidia itself...

I dont know a single business where all customers are treated equally
irrespective of the revenue they are providing , not even on the consumer
place , mind you...


As you said , if the product is successful it will be forcibly manufactured
by the two foundries , wich would be a mean for AMD to reduce the pressure from both sides thanks to the inevitable competition.

 

[CTho9305]

Intel has no competition now or in the near future. Atom will still be a joke - I guarantee you that when the 22nm Atom is compared to Kabini it will look even worse than the 45nm did vs Brazos. While it will have superior TDP characteristics, it will be because of having pathetic graphics. There is zero reason to believe anything will change from the current Atom vs Brazos situation.

Don't underestimate 22nm Atom. The microarchitecture is supposed to be massively overhauled.

If you factor the kind of market that Brazos compete and the lack of clout with OEMs, Brazos margin should be a bit lower than the average. That would put gross margin in the 40-50% range. Brazos wasn't irrelevant for the shipments there,

I think you're missing the fact that a low-price market can support good margins if your costs are low enough and your offering is enough better than any competition. Just to be clear, I'm not talking about AMD's overall margins - I'm only talking about potential margins for lower-priced markets. I don't think AMD charged enough for Bobcat early on (and evidence supporting that would be rumors that they were supply-limited at the time, despite using a mature-at-the-time, high-volume manufacturing process).

 

[pablo87]

AMD paid Global Foundries to cancel their exclusivity on 28nm. As with all things WSA the devil is in the details ( which we don't have) most important of which is, for how long?

Kumar Is a confusing public speaker so I wouldn't take what he says as absolute. He couldn't explain the discrepancy between break-even sales at $1.3B and $450MM quarterly overhead @ historical margins. Analysts were left scratching their heads.

Nonetheless, the math on taking $1.1B worth of stuff from GF in 2013 doesn't add up without Kabini's ported over and none too soon. $1.1B is what they took in 2012 of which $100MM was basically scrapped, and a whole bunch more was sitting on shelves on Sept. 30. Demand in 2013 is not likely to be any better and Kabini will soon enough be eating Trinity's lunch at OEMs.

btw, q4 earnings call sb entertaining.

 

[NostaSeronx]

AMD paid Global Foundries to cancel their exclusivity on 28nm.

For 2012.

"We are obligated to make all microprocessor products at GLOBALFOUNDRIES, and we will do that." said Devinder Kumar during the call.

What this means is that at 28nm and below, any CPU or APU will be made exclusively at GLOBALFOUNDRIES for the time being. This is unlike the present situation where AMD's low power APUs based on the Ontario and Zacate cores (Brazos platform) are manufactured at 40nm at TSMC.

http://www.brightsideofnews.com/new...gets2c-revises-globalfoundries-agreement.aspx

Separately, AMD will move to standard 28nm process technology and significantly reduce reimbursements to GF for future research and development costs.
+
R&D savings help drive AMD to its Q3’13 $450M OPEX target through adoption of 28nm standard process technology.

http://phx.corporate-ir.net/External.File?item=UGFyZW50SUQ9MTY0MjcyfENoaWxkSUQ9LTF8VHlwZT0z&t=1

For the notebook, this is not the FD-SOI standard process technology from ST-Micro rather just the Bulk from Samsung:

ASN: The plan was to start production in your fab in Crolles, then shift to GlobalFoundries for high-volume production in 2013 — is this still the schedule? From a manufacturing standpoint, what does it take to get a fab ready for FD-SOI production (does it take much longer than a typical bulk scaling transition)? Are there any special tools or other preparations needed?

JMC: For manufacturing, 28nm FD-SOI technology uses the same toolset as for 28nm LP bulk. Process development is complete, and ST/Crolles fab is now working to bring yield at production levels and complete the qualification of the technology, as done for any other.

Phase-in of the technology at GlobalFoundries is planned to start Q1 2013, with process qualified and with production level yield foreseen for Q4 2013.

http://semimd.com/hars/2012/10/08/sts-fd-soi-tech-available-to-all-through-gf/

2011 -> 32-nm
2012 -> 32-nm
2013 -> 28-nm
2014 -> 28-nm
2015 -> 20-nm

 

[mrmt]

AMD paid Global Foundries to cancel their exclusivity on 28nm. As with all things WSA the devil is in the details ( which we don't have) most important of which is, for how long?

No, they paid for a waiver. The exclusivity clause was always there.

Kumar Is a confusing public speaker so I wouldn't take what he says as absolute. He couldn't explain the discrepancy between break-even sales at $1.3B and $450MM quarterly overhead @ historical margins. Analysts were left scratching their heads.

I don't think he is confusing as much as he is inexperienced in Q&A and maybe even with financials. He seems to adhere to the script like there is no tomorrow, maybe out of nervousness or inexperience, but script is very shoddy, not really at the levels of the guys listening to the call.

Take for example the 1.3 billion snafu. It was something that a day-1 intern could listen, write in a notebook and ask the same question. Yet this question made Kumar tap dance around it twice in the Q&A.

I don't think Kumar is stupid to the point of not knowing how to calculate gross margin with the data he presented, but he is stupid for giving a go for a script that allowed him to give revenue and OPEX without giving a gross margin commitment.

That said, there is a huge difference between Kumar tap dancing around a question he isn't prepared to answer as he did with the gross margin and a few others, and Kumar outright lying like it would be if there is some kind of waiver buried into the WSA he didn't mention. I don't think he would have allowed himself to fall into the latter case.

Nonetheless, the math on taking $1.1B worth of stuff from GF in 2013 doesn't add up without Kabini's ported over and none too soon.

Agreed here. I don't see how AMD can reach their purchase commitments without bringing Kabini to GLF. There will be a lot of pressure on AMD line up once Haswell arrives, They need Kabini there to make up for the loss in volumes.

 

[Nemesis 1]

They shouldn't. GF is to blame for most of AMD's problems of the past 2 years. Beggars can't be choosers though and AMD has to make the best it can of two bad foundries.

AMD caused it not GLF It was AMD who choose the IBM path to gate first . IBM is AMDS problem . It really a funny story from amds beginnings to present day .

 

[AtenRa]

Are we talking about 17 watt processors if so . They aren't going against Intel Atom but haswell . If your talking below 10 watts . lets wait and see , Atom is at most a 10 watt chip and that I believe is server chip . If the comparison is made at same wattage I think Intel wins . But ignorer to go over the 10 watt threshold Atom isn't the compare chip Haswell is . If people are after power they will buy haswell. If they are after efficiency than Atom silvermont will win . AMD can only fit between the 2 intel products . Basicly Intel has ARM on the ropes at 22nm silvermont and. and Intel has AMD between a rock and a hard place . Only a blind man couldn't see this .

22nm, Dual Core Hasswell with GT3(40EUs) will be close to 160mm2, same as 22nm Quad Core IvyBridge with GT2(20EUs).

28nm, Quad Core, 128 SPs Kabini will be close to 90mm2, almost half the die size. I dont know why you trying to compare the two.

Kabini will not compete with 10W Hasswell due to chaotic price difference. Kabini desktop/mobile (10-25W) will compete against low price Celerons and Pentiums.
So, people that after the low price segment (the majority) they will better be with Kabini than Celerons/Pentiums.

 

[ShintaiDK]

22nm, Dual Core Hasswell with GT3(40EUs) will be close to 160mm2, same as 22nm Quad Core IvyBridge with GT2(20EUs).

28nm, Quad Core, 128 SPs Kabini will be close to 90mm2, almost half the die size. I dont know why you trying to compare the two.

Kabini will not compete with 10W Hasswell due to chaotic price difference. Kabini desktop/mobile (10-25W) will compete against low price Celerons and Pentiums.
So, people that after the low price segment (the majority) they will better be with Kabini than Celerons/Pentiums.

Why do you need a GT3 to compete with Kabini? A GT3 would be wastly superiour in both CPU and GPU performance plus lower power consumption. And we are getting sub 10W IBs now.

 

[pablo87]

For 2012.http://www.brightsideofnews.com/new...gets2c-revises-globalfoundries-agreement.aspxhttp://phx.corporate-ir.net/External.File?item=UGFyZW50SUQ9MTY0MjcyfENoaWxkSUQ9LTF8VHlwZT0z&t=1

For the notebook, this is not the FD-SOI standard process technology from ST-Micro rather just the Bulk from Samsung:http://semimd.com/hars/2012/10/08/sts-fd-soi-tech-available-to-all-through-gf/

2011 -> 32-nm
2012 -> 32-nm
2013 -> 28-nm
2014 -> 28-nm
2015 -> 20-nm

long story short, is Kabini at TSMC OR GF?

 

[pablo87]

No, they paid for a waiver. The exclusivity clause was always there.



I don't think he is confusing as much as he is inexperienced in Q&A and maybe even with financials. He seems to adhere to the script like there is no tomorrow, maybe out of nervousness or inexperience, but script is very shoddy, not really at the levels of the guys listening to the call.

Take for example the 1.3 billion snafu. It was something that a day-1 intern could listen, write in a notebook and ask the same question. Yet this question made Kumar tap dance around it twice in the Q&A.

I don't think Kumar is stupid to the point of not knowing how to calculate gross margin with the data he presented, but he is stupid for giving a go for a script that allowed him to give revenue and OPEX without giving a gross margin commitment.

That said, there is a huge difference between Kumar tap dancing around a question he isn't prepared to answer as he did with the gross margin and a few others, and Kumar outright lying like it would be if there is some kind of waiver buried into the WSA he didn't mention. I don't think he would have allowed himself to fall into the latter case.

AMD l

Agreed here. I don't see how AMD can reach their purchase commitments without bringing Kabini to GLF. There will be a lot of pressure on AMD line up once Haswell arrives, They need Kabini there to make up for the loss in volumes.

As always you make good points but I have to say, historically transparency and WSA are <>.

By the way, that waiver was expensive wasn't it? At least it was good value..

 

[Olikan]

Why do you need a GT3 to compete with Kabini? A GT3 would be wastly superiour in both CPU and GPU performance plus lower power consumption. And we are getting sub 10W IBs now.

we have clock speed on 8W Haswell?

 

[SiliconWars]

we have clock speed on 8W Haswell?

No need, intel lives in a magical land where TDP's can be halved while clock speeds and shader counts aren't affected.

 

[mrmt]

As always you make good points but I have to say, historically transparency and WSA are <>.

Lack of transparency in AMD isn't restricted to the WSA, but to every single subject you can think of. You just have to look on how few items are disclosed on their annual report. Revenue breakdown, COGS breakdown, margin forecast, margin per segment, roadmaps, all this isn't fully disclosed on AMD SEC fillings or even answered in the Q&A.

Look on how different companies deal with roadmap for example. We already know what Intel will launch in 2013, and we already know what Qualcomm will have in 2013. We had a good idea on how these products would work and the launch time frame. And they have been consistently delivering in the last few years. Now have a look at AMD. We don't know what AMD will launch in 2013 and how it will perform. Only Kabini is being disclosed, Richland, which comprises a significant part of AMD business, comparable to Brazos if you look at revenues, is a huge "?". And 2014? We already know where Intel will go, and Qualcomm, and even Nvidia for that matter. But can you say the same about AMD? No.

Same with forecasts. Intel gives revenue and margin forecasts for the entire year, AMD, pff, not even for the next quarter.

Just look on what Kumar tried to do at the last WSA Q&A. Have we been dealing with an honest and transparent company, they should step forward and mentioned that there would be no waiver and that all chips would go for GLF, but instead Kumar kept mum about the subject until directly inquired by an analyst.

I could spend the entire day recalling AMD quarterly calls and how they omitted important information from the investors, information that could change the perception from investors, but I think the point is already clear enough.

With the WSA AMD is just proceeding with their business as usual attitude, but this time what is really hurting is that the WSA is a potential liability worth many times the value of the company now. For example, how much is the new take-or-pay charge if AMD decide to not purchase chips from AMD? Are there safeguards to protect AMD from GLF delays or low yields? Are the purchase commitments a temporary relief or they are a new definitive level (I think they are not definitive, more on that later)? You can't really do a long term investing decision on AMD without answering these questions, and so far AMD still refuses to provide them.

I'm yet to see AMD cross the red line in lying straight face to investors, but the lack of transparency on these sensitive issues is almost on par with that.

By the way, that waiver was expensive wasn't it? At least it was good value..

You know, one of the jokes in the O&G sector is that the most profitable business of the world is a poorly managed oil company. I disagree. The most profitable business of the world is AMD foundry. No matter how much you screw up, your customer will still pay you a lot of money.

 

[Nemesis 1]

No need, intel lives in a magical land where TDP's can be halved while clock speeds and shader counts aren't affected.

Not one person here has said they would be . You clearly see this . Yet you ignor the fact lowpower IB are better in graphics than low power trinity . You make a point of pointing this fact out about intel cpus . Like its a intel exclusive . Its not and were not stupid

 

[Olikan]

No need, intel lives in a magical land where TDP's can be halved while clock speeds and shader counts aren't affected.

well, i do expect 8W Haswell beat kabini... but not at a large margin, and amd catching up in multi-threaded...

the IGP war will be really intense with GT2 vs GCN

 

[mrmt]

the IGP war will be really intense with GT2 vs GCN

Just to show how AMD is being "outinvested" in the CPU/IGP war. Three years ago when Westmere launched Intel graphics were barely enough to run Windows. Now in 2013 people are discussing if Haswell will beat Trinity and in which steps on the power ladder.

Atom is going for the same release cadence as Core and will have a lot of features in common with Core. If someone still thinks that Kabini can hold its own with an slower launch cadence and a process handicap, it is better to start review these premisses.

 

[AtenRa]

Why do you need a GT3 to compete with Kabini? A GT3 would be wastly superiour in both CPU and GPU performance plus lower power consumption. And we are getting sub 10W IBs now.

According to the slide bellow, 10W TDP Hasswell is only GT3. Nemesis1 by referring to 10W Hasswell was pointing to dual Core GT3. This CPU will be a very expensive ULV and not in the same price point of Kabini.

And we are getting sub 10W IBs now.

Any links ?? because the lowest i could find is the Pentium 2117U at 17W(22nm).

http://ark.intel.com/products/71469/Intel-Pentium-2117U-Processor-2M-Cache-1_80-GHz

If Pentium 2117U, a Dual Core Dual Threads 1.8GHz with HD graphics is at 17W, a 10W Hasswell with GT3 will be at 1-1.2GHz. The GT3 will be underutilized most of the time because of the low TDP limit.

 

[Nemesis 1]

Here is a semi a, article using the slides used in here to report the delay of Silvermont merrifield phone chip . its from may 11th. 2012 .
No were I read from intel that merrifield is delayed . Last jan. Anand did medfield article . READ it . He hints that merrifield is a 1st qt 2013 part . None of that has changed . The only thing that has changed is airmont on 14nm . The tablet version of Silvermont comes late in 2013 early 2014 . Nothing has changed . Other than 14nm . The recent merrifield bench as linked shows it be 4g not 2g .
z580 is a tablet chip recently released . If Intel can put that chip in a phone , Thats a bigger feat than having 22nm merrifield ready for 1st qt .
One simplely needs to google merrifield . I find it a bit hard to believe the 544 @533 capped out at 60fps . I don't believe its that good . We just have to wait and see . The rooster will cock off. The crows will eat crow . After thinking on it If intel can get Z2580 into a phone . Thats actually exciting and surprizing I don't believe intel is that good

 

[Nemesis 1]

According to the slide bellow, 10W TDP Hasswell is only GT3. Nemesis1 by referring to 10W Hasswell was pointing to dual Core GT3. This CPU will be a very expensive ULV and not in the same price point of Kabini.

Any links ?? because the lowest i could find is the Pentium 2117U at 17W(22nm).

http://ark.intel.com/products/71469/Intel-Pentium-2117U-Processor-2M-Cache-1_80-GHz

If Pentium 2117U, a Dual Core Dual Threads 1.8GHz with HD graphics is at 17W, a 10W Hasswell with GT3 will be at 1-1.2GHz. The GT3 will be underutilized most of the time because of the low TDP limit.

That true thats with the haswell 8 watt unit . Ivb is at 10watts without soix and is not SoC as the haswell tablet chip . But in referance to silvermont . I don't know whats in it for gpu in merrifield . But the tablet gets IB hd4000 eu . AT i believe said that it would be 4 eu . I don't believe that unless 4 eu phone merrifield. Why would intel cripple GPU performance in a tablet . What GPU is in the 10 watt IB. IN the tablet version of IB what is the eu count . In haswell 8 watt . what is the EU count 40eu . Seems the 4eu is a joke in a tablet silvermont . But in merrifield that would be a great gpu . I expect the HD 4000 to be in the tablet silvermont 20 eus should be easy to stay in power range . We have IB @10 watts We will have haswell gt3 in the 8 watt haswell . So why would intel put a 4 eu gpu in silvermont tablet running 4 non ht cores . ANS they wouldn't

 

[mrmt]

According to the slide bellow, 10W TDP Hasswell is only GT3. Nemesis1 by referring to 10W Hasswell was pointing to dual Core GT3. This CPU will be a very expensive ULV and not in the same price point of Kabini.

What about those 13.5W GT2 part on the slide?

 

[SiliconWars]

Not one person here has said they would be . You clearly see this . Yet you ignor the fact lowpower IB are better in graphics than low power trinity . You make a point of pointing this fact out about intel cpus . Like its a intel exclusive . Its not and were not stupid

No they aren't. You might be able to find a few cpu-bound titles where intel wins but in most games they get thrashed vs Trinity at the same TDP.

Did you see the diablo III results with the 35W Trinity vs the 55W HD4000?

THRASHED. Even the 17W Trinity is faster than the 55W, and this result will be the same in many other games.

Just to show how AMD is being "outinvested" in the CPU/IGP war. Three years ago when Westmere launched Intel graphics were barely enough to run Windows. Now in 2013 people are discussing if Haswell will beat Trinity and in which steps on the power ladder.

Considering intel will have more power budget and transistors thrown at gpu than ever before, it's not exactly surprising. Not to mention more money thrown at graphics R&D than the rest of the entire industry combined. Give them a huge round of applause for having more money than everybody else due to being there first, and finally getting on graphical parity years later. Great stuff!

 

[SiliconWars]

http://www.anandtech.com/show/6194/8

Intel cannot get their graphics to work at full tilt at 17W, they have no chance of getting any kind of performance at 8W on the same node. You're looking at sub 300 MHz clocks, no turbo. Kabini will thrash this, in fact I'll be surprised if the current C-60 Bobcat wouldn't thrash it. The x86 cores are too powerful and can't scale down so far without serious compromises being made.

Intel's desperation to get IVB and Haswell below 10W is simply because Atom sucks and nobody wants it in tablets. Kabini and Temash will be so much better because they have been aimed properly at this segment while intel attempts to blow up a phone chip and scale down a server chip to fit.