Extremetech: "AMD cancels 28nm APUs, starts from scratch at TSMC"

[drizek]

What did Apple drop AMD from?

Next gen MBPs to have nvidia GPUs. Macbook Air was built with Llano but they picked SB instead.

 

[drizek]

That has got to be one beauty of a screen with that ppi.

The iphone is 320

I think 280 or so should be the target, the iphone is a bit overkill, but it is very very nice.

 

[Phynaz]

Next gen MBPs to have nvidia GPUs. Macbook Air was built with Llano but they picked SB instead.

Apple has never used an AMD cpu.

 

[Arkaign]

That has got to be one beauty of a screen with that ppi.

It's not quite on the iPhone 326ppi 'retina' display's level, but yeah, that's pretty darned decent.

 

[lifeblood]

Really , It was AMD who choose gate first. Amd made many decisions befor selling their fabs . It looks to me like AMD should have never bought ATI . 5+ billion for a gaphics company that was worth 2 billion . AMD has made one bad decision after another. They could have developed their own graphics and been way ahead of the game . Now they are screwing GF because of AMDs decision making.

AMD is not completly innocent, but they are not as guilty as you make them out to be. Making their own graphics from scratch and having it be competitive with ATIs and nVidias is not as easy as you make it sound. AMDs decision to buy ATI was a good one, although the price they paid for them was pretty bad (no putting lipstick on that pig). LLano has been a success but a limited one because of GFs inability to ramp up production. Bulldozer could have been a success had it come out sooner and hit higher clock speeds, but it failed on both counts primarily because of GF.

As far as CPUs are concerned, AMD has been the innovator, not Intel. Intel's been able to sit back and be content to follow along (innovation wise, not performance wise) because they have the superior fabs and processes.

 

[sandorski]

Clearly it is getting increasingly difficult to move Processes lower and lower. The last couple years has seen too many Delayed and even Cancelled Products due to the lack of Manufacturing capability. It really makes me wonder if AMD and others have reached a wall and only Intel is capable of avoiding it?(technically already has)

As for GF and AMD shouldn't have let them go? I'm thinking that letting GF go may have been the wise, rather than foolish, move. There's no guarantee that AMD would have succeeded where GF has failed, I suspect AMD saw this risk and that's why they bailed in the first place. If they still controlled the Fabs, they'd probably be in more trouble now than they currently are.

 

[Ajay]

Next gen MBPs to have nvidia GPUs. Macbook Air was built with Llano but they picked SB instead.

Supposedly, nVidia got the design wins, and is getting allot wins with it's mobile Kepler parts (paired with IB CPUs).

 

[thilanliyan]

That has got to be one beauty of a screen with that ppi.

Yep I wonder when we will get something like that in North America...

I'm looking to upgrade my Nook Color (1st gen) and almost bought a Playbook with the recent "firesale"...but am holding off for now.

 

[thetuna]

I thought Apple excommunicated Nvidia or something, no?

 

[cbn]

With the "Gate First" problem for 28nm Bobcat APUs, I just wonder if this could be the straw that finally broke AMD's commitment to x86?

ARMv8 for AMD?

 

[LoneNinja]

Yeah, I don't really understand that post either.

There were rumors early this year that Apple was going to use Llano in some of it's upcoming products, and later when it didn't happen it was followed by rumor that AMD couldn't guarantee Apple a high enough quantity of Llano chips so they stuck with Intel and Sandy Bridge.

 

[cbn]

There were rumors early this year that Apple was going to use Llano in some of it's upcoming products, and later when it didn't happen it was followed by rumor that AMD couldn't guarantee Apple a high enough quantity of Llano chips so they stuck with Intel and Sandy Bridge.

Yep, Here is the article from S/A making the claim.

P.S. [In the third paragraph of the above article] How do people feel about Charlie's sources claiming Apple is still giving the nod to ARM 64 bit in 2013? (Maybe an ARM Mac Book Air combined with high resolution Mirasol display to accentuate the power savings of the SOC?)

 

[ocre]

With the "Gate First" problem for 28nm Bobcat APUs, I just wonder if this could be the straw that finally broke AMD's commitment to x86?

ARMv8 for AMD?

Not hardly. If AMD decided to enter ARM now, they arent gonna be able to have products for some time. They would still be committed to x86. It would take them a couple yrs to ready an ARM entry. Most likely (and more wisely), they would try hard to keep a plan like this a secret as long as possible.

Eventually i think the interest in ARM is inevitable. But i believe AMD will not push full force on such a plan. I see them walking the middle here. Meaning they will half heartily put resources in an ARM package while still focusing on x86. That is if they even decide to go with ARM at all. Arm has an incredible amount of competitors to which AMD is years behind already. They can offer little advantage years down the road. While many ppl will claim GPU, at this time they are using very powerful GPUs for the very low power limits they work in. If arm scales up, perhaps AMD has a gpu future there. They are trying to make good ARM relations and its not a bad idea.

I believe AMD has lost its way. I see them as sitting on the fence, currently. They dont seem to be taking any big risk while the stuff that shouldnt be to riskful is all backfiring in their face. I see their new strategy as a belt tightening strategy not an abandon x86 one. If intel can manage to bring back the x86 fire, AMD is in a better position than it would be in ARM. If windows 8 backfires and benefits x86 more than it does ARM, than AMD is in a much better position with x86. I am sure AMD is on the fence waiting for a clear road, they arent about to throw all their chips at anything. This is the AMD i see.

If AMD has any ARM desires, it will be a very half azz program that will be kept away from the public. With their exclusive membership to x86, its in their best interest to stay with it or at least appear to completely have confidence in x86. At the same time, if they are making a good relationship with ARM it could benefit them long term.

I believe AMD has technologies it will sell, patents and licenses which it could use to make some much needed cash. These technologies also build a relationship which could pay off well. I dont see how its right for AMD to drop x86 and go ARM. Win8 could very well give AMD a great advantage with their fusion chips in tablets. It makes zero sense for them to publicly break their x86 commitment at this time.

 

[RussianSensation]

It appears the main issue of Apple ditching AMD is not a technical aspect of their GPUs.

http://blogs.barrons.com/techtraderdaily/2011/11/21/nvda-lonbow-agrees-they-pushed-out-amd-at-apple/

It could be Reed isn't interested in changing a roadmap to appease Apple -- it might not be financially/strategically worthwhile to divert limited resources just to make a extra $40-45M in revenue. AMD would be better off focusing its resources on profitable emerging markets, enterprise customers and building a stronger discrete GPU business with higher average selling prices. There could be other reasons such as NV outbidding AMD this round.

 

[cbn]

If AMD decided to enter ARM now, they arent gonna be able to have products for some time. They would still be committed to x86. It would take them a couple yrs to ready an ARM entry. Most likely (and more wisely), they would try hard to keep a plan like this a secret as long as possible.

Agreed. I think AMD would need to keep selling 40nm Bobcat APUs for as long as it could. Keeping ARM plans a secret makes total sense too.

If intel can manage to bring back the x86 fire, AMD is in a better position than it would be in ARM.

I am sure Intel will make some great x86 chips. Their process technology (Tri-gate and successors) will assure this. But does this really help AMD? The way I see things now Intel's efficiency encourages Microsoft "Big Windows" to become even lazier (which hurts the performance of the entry level x86 offerings from a software standpoint)

I can almost see the Microsoft CEO saying "Wow, look how cheap and fast these Intel CPUs are getting! Let's cut the x86 OS software budget to increase profits! Instead, we will focus our programming efforts into making the ARM Phones run faster."

In a nutshell:

Process technology for performance per watt (Intel)

vs.

Software writing for performance per watt (Mobile OS/ARM)

Where would AMD fit within these extremes?

 

[wlee15]

I wonder if elements of Bobcat get merged into to the mainstream designs; there's quite a bit of interesting elements such as the low power floating point multiplier that actually outperforms additions in scalar 32-bit (and indeed every other x86 floating point multiplier).

 

[zlejedi]

This bobcat news is worse than I thought....

but maybe the low power display techs will come to the rescue of the lowest power x86 laptop chips? (re: low power displays help accentuate any difference in power consumption between platforms)

How would it help 40nm Bobcat competing against 22nm Ivy Bridge ULV chips ?

 

[MisterMac]

Like others say:


Doesn't the majorly atleast, most of the market wait for how Windows 8 will run and work, and especially work with ARM?

ULV nettops and tablets will undoubtly flood the market, should W8 run well on it.

Eventually it will run obviously, but if the software budget is still mainly devoted to x86 and it's just microsoft tippy toeing for new markets - AMD keeps being in a screwy position.

Assume they're secretly working on a ARM APU/CPU, they sort of need windows to push the market in that direction and be able to enter a market where one competitor doesn't have a retarded unfair process advantage ?

That seems the logical step for me.

AMD won't be able to make x86 ULV chips, with enough performance, especially comparing to Intel.

So they sort of need ARM to be pushed into that area by microsoft, don't they?

 

[AtenRa]

Can anyone point me to a link saying AMD was going to use ONLY GloFos 28nm for Bobcat??

As far as i remember, AMD was going to use TSMC's 28nm for Ontario/Zacate replacements.
It was the reasonable choice because first gen Bobcat used TSMCs 40nm process, the jump to 28nm would be easier dealing with the same Fab(not the only reason).
Secondly, it was known more than a year ago, that GloFos 28nm at Fab 1 in Dresden was going to be production ready after TSMCs 28nm .

AMD had the choice of two 28nm process Fabs, GloFo and TSMC. Im sure they (AMD) evaluated both processes and chose TSMC a long time ago, keeping GloFo for backup maybe.

It seams to me that the article is talking about the scraping of Bobcat 28nm GloFo's pack up plans because TSMC's 28nm will be ready for Q1 2012.

 

[ocre]

Agreed. I think AMD would need to keep selling 40nm Bobcat APUs for as long as it could. Keeping ARM plans a secret makes total sense too.



I am sure Intel will make some great x86 chips. Their process technology (Tri-gate and successors) will assure this. But does this really help AMD? The way I see things now Intel's efficiency encourages Microsoft "Big Windows" to become even lazier (which hurts the performance of the entry level x86 offerings from a software standpoint)

I can almost see the Microsoft CEO saying "Wow, look how cheap and fast these Intel CPUs are getting! Let's cut the x86 OS software budget to increase profits! Instead, we will focus our programming efforts into making the ARM Phones run faster."

In a nutshell:

Process technology for performance per watt (Intel)

vs.

Software writing for performance per watt (Mobile OS/ARM)

Where would AMD fit within these extremes?

AMD still sells more CPUs than nvidia. Nvidia has been active in the ARM space for a good while now. You think AMD could immediately jump to ARM and sell enough chips to keep them from collapsing? ARM margins arent as high as x86. AMD will not be the performer nvidia has in the new space, they are yrs behind. Its something that will take some time, and AMD is a huge company. huge!

Their best bet is to stick with x86, where they still sell more CPUs than nvidia. They are capable of selling chips here, right now! They cant jump to ARM without a massive bump. A massive bump they dont have the funds to withstand.

I know it may not sound like x86 is their best bet, but it really is. Their chips do everything intels do. They arent as fast, but from a consumer standpoint they function just as well in every day task. AMD is making profits. Remember AMD existed for yrs in the x86 market without any profits. Years!

The ARM market is flooded with competition. I would not doubt on AMD having a half azz interest, a back up exit strategy. If they dont, they need to. But its in their best interest for x86 to gain momentum. They are gonna be eating intel crumbs for a long while. They cannot afford to drop x86 any time soon, the detour will not be smooth and they are much further behind in ARM technology than they are in x86.

 

[IntelUser2000]

It doesn't make sense. So assume three options AMD had:

-Continue with TSMC on Brazos successor
-Switch to Global Foundries exclusively for Brazos successor
-Dual source GF and TSMC

The least likely is number 3. Not only they'll have to manage two suppliers but create entirely different dies for each company as well. Gate First vs Gate Last only adds to the argument pie.

Assume #1, two possibilities:
-Extremetech and SA are both attention whores, move along
-AMD is going to move to TSMC across the entire product line. From Servers to Ultra mobile and CPUs to GPUs. The possibility is there, server roadmaps are out there that says the road after 32nm is 28nm

Assume #2: Similar to #1 but with the difference I'll give press a nod for actually being correct.

 

[theeedude]

Moving 28nm APUs from GlobalFoundries to TSMC means scrapping the existing designs and laying out new parts using gate-last rather than gate-first manufacturing. AMD may try to mitigate the damage by doing a straightforward 28nm die shrink of existing Ontario/Zacate products. While the resulting chips wouldn’t incorporate any of the architectural improvements originally intended for Krishna/Wichita, they’d lower Brazos’ power consumption and buy time for a new processor.

This does not make sense to me. If it's only an issue of GF yields, it should be easier to port an existing 28nm GF design to TSMC 28nm libraries than it is to scale a 40nm design to 28nm, and keep all the architectural improvements. There is no such thing as straightforward shrink anymore, design rules don't all scale, you have to redo the layout anyways. If they are scrapping their new 28nm design, there is probably more wrong with it than just GF yields.

 

[Ajay]

-AMD is going to move to TSMC across the entire product line. From Servers to Ultra mobile and CPUs to GPUs. The possibility is there, server roadmaps are out there that says the road after 32nm is 28nm

Do either TSMC or GF have a 28nm SHP node?

 

[bryanW1995]

AMD is not completly innocent, but they are not as guilty as you make them out to be. Making their own graphics from scratch and having it be competitive with ATIs and nVidias is not as easy as you make it sound. AMDs decision to buy ATI was a good one, although the price they paid for them was pretty bad (no putting lipstick on that pig). LLano has been a success but a limited one because of GFs inability to ramp up production. Bulldozer could have been a success had it come out sooner and hit higher clock speeds, but it failed on both counts primarily because of GF.

As far as CPUs are concerned, AMD has been the innovator, not Intel. Intel's been able to sit back and be content to follow along (innovation wise, not performance wise) because they have the superior fabs and processes.

I think that your scale is off. My decision to stay out an extra 90 minutes the other night and drink 4 more beers was pretty bad because it gave me a little bit more of a hangover the next day. Paying $5 billion for ATI was more like staying out all night, hooking up with a stripper, giving her your home number by mistake, wrecking your car on the way home, losing your job, and getting divorced (in California) when the stripper wouldn't stop calling your house.

 

[lifeblood]

I think that your scale is off. My decision to stay out an extra 90 minutes the other night and drink 4 more beers was pretty bad because it gave me a little bit more of a hangover the next day. Paying $5 billion for ATI was more like staying out all night, hooking up with a stripper, giving her your home number by mistake, wrecking your car on the way home, losing your job, and getting divorced (in California) when the stripper wouldn't stop calling your house.

Now that's harsh.

Not that I'm saying its not true...