AMD cuts to the core with 'Bulldozer' Opterons

[flippin_waffles]

WOW! So GF giving you guys the chips for free? I know you want to make it sound like it's all positive to go fabless but reality is very different since you are paying per wafer, die whatever and at a premium and GF will now have their own priorities that may not go inline with Amd's so they could push out process shrinks out if they see it in their best interest. I have to wonder how much you are really saving by losing control of your manufacturing and I would love to hear what your old CEO/founder Mr. Sanders feels about Amd's new fabless model.

I don't think AMD will be paying the equivilant of the $4-5 Billion on wafers, that it costs to build and equip a fab and develop a new process.

What i'm wondering, is what intel plans to do with all their under utilized fabs and employees if and when they fall below %70 market share, and margins start to drop along with them. AMD is much leaner and already at a point where they can operate on much lower margins. intel may have to make some adjustments in their business model also. I'm sure they're preparing to avoid that with whatever means necessary. Here's a few past examples of how they might accomplish that, taken from another forum:

I'm curious as to what the rules are... Are they written down somewhere? Because by my own ethics meter, I dislike Intel precisely because of the shady crap they've pulled over the 25 or so years I've been in the inudstry. I went from proudly buying my then top-of-the-line 80486DX (50mhz!!) to generally not trusting and at times loathing them.

-The FDIV controversy. If your product is broken, offer to fix it, without the initial public statements of, "You have to prove your CPU is broken and that you actually use your FPU before we'll fix it." (Yes, they backed off on this, but if they'd done the 'right' thing to start with, then we'd be talking about it as a shining example of a company doing the honorable thing rather than the inverse, 16 years later.
-Threatening MB and chipset makers to not use a technically superior product (I'm talking about the K7 launch and the fact that there MB manufacturers were afraid to show a product supporting the K7, and the ones they did did so without markings on the box
-Doing a demo of a 3D game with one of the first dedicated 3D processors (3DFX in this case), which was responsible for a dramatic improvement in the gaming experience, and suggesting via demo that the difference was the anemic (at the time) CPU extensions
-Bribing manufacturers to not use a superior part from a competitor (Dell payoffs, Opteron)
-Changing a benchmark suite by removing the portions of the benchmark that your competitor wins (sysmark 2000 and 2001 fiascos)
-Refusing to use supported extensions on competitor's CPUs

The problem with consumers is, we have a very long memory when it comes to things like this. To this day I still don't use TurboTax, after that DRM stunt they pulled a few years back.

to name a few

http://www.semiaccurate.com/forums/s...t=1226&page=27

[qurious63ss]

I don't think AMD will be paying the equivilant of the $4-5 Billion on wafers, that it costs to build and equip a fab and develop a new process.

What i'm wondering, is what intel plans to do with all their under utilized fabs and employees if and when they fall below %70 market share, and margins start to drop along with them. AMD is much leaner and already at a point where they can operate on much lower margins. intel may have to make some adjustments in their business model also. I'm sure they're preparing to avoid that with whatever means necessary. Here's a few past examples of how they might accomplish that, taken from another forum:



to name a few

http://www.semiaccurate.com/forums/s...t=1226&page=27

They will do what every company does when they don't need employees or factories layoffs and closures Now will they go fabless? Probably not.

[informal]

qurious63ss you conveniently avoided my post

[qurious63ss]

qurious63ss you conveniently avoided my post

Sorry, didn't realize it was directed towards me. Yes you make some good points and from an ideal Amd world then it appears that it is a win win for Amd to go fabless, but the things I brought up are possible to no? My point has been that going fabless has its drawbacks, time will tell who is right.

[JF-AMD]

When paying by the wafer vs. buying a fab, by the wafer is higher. However, unless you are running at 100% capacity, you have high sunk costs that can be harder to recoup.

The question is do you believe there is better flexibility in higher fixed costs or higher variable costs. Believe it or not, especially in today's crazy market, there is more flexibility in a variable cost model. Just look at how much debt AMD was able to buy down - FAB = debt and debt chokes off profit. I'd rather hold the cash and direct it at more wafers if I need it.

[qurious63ss]

When paying by the wafer vs. buying a fab, by the wafer is higher. However, unless you are running at 100% capacity, you have high sunk costs that can be harder to recoup.

The question is do you believe there is better flexibility in higher fixed costs or higher variable costs. Believe it or not, especially in today's crazy market, there is more flexibility in a variable cost model. Just look at how much debt AMD was able to buy down - FAB = debt and debt chokes off profit. I'd rather hold the cash and direct it at more wafers if I need it.

I agree that variable cost is a good thing, but do you agree that giving up manufacturing of your products, especially one that relies on leading edge tech like processors, has some negative impacts since you loose control of the process and are relying on another company to stay on schedule all this while they are looking out for their best interest before yours. Lets put it this way if the cost to fab in house was the same as to foundry which would you go with? The way I see it, it's a gamble that all boils down to money. You are willing to give up the process in order to save money, while Intel is willing to spend the money in order to control the process. I see no problem with your model when it involves low margin product like flash, chipsets etc. but when it involves processors I would definitely want control of the process to synch with my products.

[JF-AMD]

I agree that variable cost is a good thing, but do you agree that giving up manufacturing of your products, especially one that relies on leading edge tech like processors, has some negative impacts since you loose control of the process and are relying on another company to stay on schedule all this while they are looking out for their best interest before yours. Lets put it this way if the cost to fab in house was the same as to foundry which would you go with? The way I see it, it's a gamble that all boils down to money. You are willing to give up the process in order to save money, while Intel is willing to spend the money in order to control the process. I see no problem with your model when it involves low margin product like flash, chipsets etc. but when it involves processors I would definitely want control of the process to synch with my products.

Yes, every tansaction has multiple impacts.

When you consider the node/design world, decisions about node and design happen years apart and rely on each other. If GF had 22nm tomorrow, unless their customers were planning on availability on that date, the designs would not be ready. The tree falls in the forest and nobody hears it.

The reality is that these things run together. Generally speaking, GF's roadmap is the same as the old AMD process roadmap would have been. Based on the fact that they are actively selling capacity, they actually have an incentive to get to the node early. If it was only AMD, I would think that GF would line up around AMD's dates, not push to get there earlier.

[Nedjo]

I have to wonder how much you are really saving by losing control of your manufacturing and I would love to hear what your old CEO/founder Mr. Sanders feels about Amd's new fabless model.

go on and search some of my posts from the last year, and you'll find quotes from the interview that Sanders gave many years ago! In that interview he has foreseen this future for AMD and has hand picked Hector to make this future come true!

Not trying to discredit Mr. Fruhue, I'm just providing a counter point to his win win comment.

that's just a point - you can't counter his point unless you have knowledge of an AMD or GloFo insider! What you're writing is based purely on your fact-less subjective POV and can't be used to counter opinions of those who have reliable info of the system with in!

[haylui]

And here is another problem. How much is Amd going to have to pay for what GF is calling "custom" manufacturing?

http://globalfoundries.com/technolog...nced_tech.aspx

As you see from the link, it looks like they are going to be doing a lot of processes at the same time.

so long GloFo could lead in technology term for contract manufacturing, they won't be mind to do so;
Don't forget they would like to have 50% of the market share in near future.
With advanced manufacturing technology, ARM and other power concern chips would be build out of the latest wafer tech rather than cheapo old tech that consume a lot of power.
More over, AMD didn't wish to withdraw from GloFo completely, but when ATIC injecting more cash into GloFo, AMD just shrink their share percentage only. Nothing to lose.

[Opteron146]

The reality is that these things run together. Generally speaking, GF's roadmap is the same as the old AMD process roadmap would have been. Based on the fact that they are actively selling capacity, they actually have an incentive to get to the node early. If it was only AMD, I would think that GF would line up around AMD's dates, not push to get there earlier.

Hmmm according to an analysts' statement in the eetimes, these things are adversaries:

Penn pointed out that outsourcing chip companies will, at the very least see wafer prices increasing. At worst they may miss market windows for lack of chips. "In many ways allocation is a bigger problem than prices increasing," said Penn.
The background to Penn's commentary is that, with a lack of investment in manufacturing capacity over the last three years, ASPs are set to rise at the same time as the general economy is recovering.
(...)
With ASPs falling at 2.8 percent CAGR from 2005 to 2009 and the overall chip market effectively stalled between 2004 and 2009 (just 1.1 percent growth) the temptation to slough off high capital costs of manufacturing is easy to see, said Penn. With cheap foundry prices and fabless companies growing fast it was easy to take the view that a fab has no market value and that manufacturing is a sevice that can be outsourced just like test and assembly.

But that is wrong, Penn said. It only holds true when manufacturing capacity is over-supplied.

Many companies will flatter themselves that they have a good relationship with their outsource supplier, but not everyone can be an A-list customer. "If you are not an A-list celebrity you could be delayed access to technology by six months. That could be the difference between success and decay."

Penn favors IDMs that continue to build fabs but use foundries to smooth the cyclic peaks and troughs in demand. The problem is that such models are not in the foundries best interest who see orders being pulled back into the IDM "That's a classic adversarial them and us business model. Clearly not in the foundries' best interests, he said. "The problem is that the investors and bean counters believe it is cheaper to outsource than to build wafers in house."

http://www.eetimes.com/news/latest/s...2600264&pgno=1

The roadmaps might fall together for now, because the new GF roadmap is identical with the old AMD FAB roadmap, furthermore GF is one of the few companies which is building new Fabs (the one in Malta,NY) but again - this is due to the old AMD plans. In the (long) term future, if the connection to AMD is history, there are obvious risks ...

If I would be AMD, I would not sell off all of my remaining GF stocks, just to secure an "A-celebrity" status ;-).

[Sn0wm@n]

loll all of you guys are arguying with jf-amd the man who's in constant touch with server grade hardware... product prices etc... basicly how to be competitive in a market that deffines the rest of your brand.... this man knows more then you think... give him credit....

[nn_step]

Yes, every tansaction has multiple impacts.

When you consider the node/design world, decisions about node and design happen years apart and rely on each other. If GF had 22nm tomorrow, unless their customers were planning on availability on that date, the designs would not be ready. The tree falls in the forest and nobody hears it.

The reality is that these things run together. Generally speaking, GF's roadmap is the same as the old AMD process roadmap would have been. Based on the fact that they are actively selling capacity, they actually have an incentive to get to the node early. If it was only AMD, I would think that GF would line up around AMD's dates, not push to get there earlier.

That of course is assuming a certain level of process optimization. Using just a design written in hardware description language like such as VHDL or Verilog, can quickly be ported to the new process. [Admittedly at lower performance than a fully optimized design] Shouldn't logically cost reduction and acceptable performance be the primary targets in the modern desktop/laptop environment rather than performance; which largely has its market in the server world which historically tends to be willing to accept a higher price and a better tuned chip on an older process?

[JF-AMD]

The reality is that interim products are not really cost effective. Let's say that GF suddenly called and said "hey, we can get to 32nm in June." If I have Magny Cours coming out in Q1, and Bulldozer coming out in 2011, trying to do an MC design in the back half of the year on 32nm would net a lot of cost and an unoptimized MC. There might be a small benefit, but by the time I could get the product to market, Bulldozer would probably be a quarter away.

OEM partners would not want to do the full validation (and they do that for every node change) so they would pass on the part.

There just isn't benefit, the most profitable answer is to keep to the roadmap and deliver to that, otherwise you impact everything trying to jam an extra project in.

[nn_step]

The reality is that interim products are not really cost effective. Let's say that GF suddenly called and said "hey, we can get to 32nm in June." If I have Magny Cours coming out in Q1, and Bulldozer coming out in 2011, trying to do an MC design in the back half of the year on 32nm would net a lot of cost and an unoptimized MC. There might be a small benefit, but by the time I could get the product to market, Bulldozer would probably be a quarter away.

OEM partners would not want to do the full validation (and they do that for every node change) so they would pass on the part.

There just isn't benefit, the most profitable answer is to keep to the roadmap and deliver to that, otherwise you impact everything trying to jam an extra project in.

Perhaps there is a misunderstanding in what was said.
I am suggesting producing an already validated design [Such as current or previous generation design] on a new process node to reduce production costs in the desktop market.
Unless of course you are suggesting that a design written in VHDL or Verilog, does not function as specified.

[god_43]

you know i just tried to find any type of quote or news by the former ceo jerry sanders......either this guy has died, or he does not talk at all.

[JumpingJack]

you know i just tried to find any type of quote or news by the former ceo jerry sanders......either this guy has died, or he does not talk at all.

Here's a recent one:
http://www.dailytech.com/Former+AMD+...ticle16694.htm

Jerry Sanders, AMD's co-founder and CEO seemed to blast Mr. Ruiz in a recent interview, stating, "It just doesn't make sense. People make dumb mistakes — talk show hosts having sex with subordinates, at least I understand the sex drive. I don't understand this. You just don't talk about things that aren't public. You don't talk to people about insider information, whether you benefit from it or not."

He always is good for speaking his mind.

[JF-AMD]

Perhaps there is a misunderstanding in what was said.
I am suggesting producing an already validated design [Such as current or previous generation design] on a new process node to reduce production costs in the desktop market.
Unless of course you are suggesting that a design written in VHDL or Verilog, does not function as specified.

My point is that with tight resources, you don't do 2 different designs, just to be ready, in case they are ready early. You have to pick your targets and run with them.

I'd rather have resources working on my next generation project than doubling down on an existing design. It's just about long term vs. short term. But, then again, I am not a desinger, I focus more on the market and I know that putting too much churn on my partners for short term qualifies then they are less likely to want to pick up my products in their lines.

[Helmore]

With Amd no longer having a say in GF decisions then there is the possibility that GF could change their focus away from Amd no matter how much Amd keeps GF full, GF will still make decisions based on whats best for GF. Some of their interest will be the same like full capacity but others like shrink cycles and the cost that comes with it might not be so. Again, not saying that this will happen but loosing control of manufacturing has its negatives and this is one of them.

I agree with you and not with just this post here.
But you have to consider that chances of that happening are reasonably slim in this case. It would be suicide for GF to slack off on process technology innovations at this point. Just consider the situation they're in. They do have a reasonable customer base, although that's only through their merger with Chartered Semiconductors, but they have no proven track record in the fab industry. This currently gives them only one option and that's to prove to their clients that they have the superior process. They need a better process than their competitors and they have to be faster than their competitors in getting large volumes and good yields on this cutting edge process. That's what they currently need to prove themselves in the fab industry.

Then later on they can choose to do what you're saying, namely to slow down on introducing smaller process nodes, but even then it will still be better for them to keep on innovating. That's because margins are much higher in the more recent process nodes than in the older nodes, mainly because there is less competition in the newer nodes as fewer fabs will have mastered it. You might even be the only one for a little while with a newer node.

Then there is one market where there will probably be quite some growth in the coming years and thats in smartphones. This will mean even more chips on the newer process nodes. Just look at Tegra 2 and future Qualcomm SnapDragon chip, all on TSMC's 40 nm node. There will be many many more of these chips in the future and they will all want to use the latest process technologies as that gives them a speed and power consumption advantage. That's another market that's very important for GF and TSMC, one in which they will have to prove which is faster in ramping up new nodes and which has the better performing process.

[JF-AMD]

I can't speak for GF because I don't work for them, but think about the remoteness of the option to move away from AMD.

AMD is their largest customer. You make money in a fab business by keeping the fab running at 100% at all times. That allows you to depreciate the fab quicker and get on to the next process.

Who, exactly, would they replace AMD with? Who else in the business could replace that much business that quickly? If they moved away from AMD, they would have a very expensive fab sitting at low utilization and that would be very costly. GF and AMD have a vested interest in working together. All of the people thinking that somehow this deal is bad for AMD don't really understand how the financial aspects of running a fab work.

Fabs are most profitable when they are at 100% utilization. They have massively high fixed costs, and relatively low variable costs. So the quicker you can run through enough wafers to offset the fixed cost, the quicker you can get to the next node. By having AMD and other companies together GF can keep the fabs running around the clock, diversify their process technologies and increase their access to IP. All of that makes them more competitive.

Dropping customers, running low utilization and starting/stopping the lines will make the fab less profitable. There is a finite number of companies that you can strike a foundry deal with, so the last thing you want to do is lose one of the big dogs.

[Nedjo]

you know i just tried to find any type of quote or news by the former ceo jerry sanders......either this guy has died, or he does not talk at all.

http://silicongenesis.stanford.edu/t...ts/sanders.htm

[JF-AMD]

Jerry was just out at the austin campus a few months ago to celebrate the 40th anniversary of AMD

[nn_step]

My point is that with tight resources, you don't do 2 different designs, just to be ready, in case they are ready early. You have to pick your targets and run with them.

I'd rather have resources working on my next generation project than doubling down on an existing design. It's just about long term vs. short term. But, then again, I am not a designer, I focus more on the market and I know that putting too much churn on my partners for short term qualifies then they are less likely to want to pick up my products in their lines.

if I am understanding this correctly, drop in replacements would cause churn for the partners and 2 engineers responsible for porting a previous design to a new process is too taxing on resources?

I may be completely mistaken in my view and perhaps am missing a few details; but I've seen students implement microprocessor designs in several different processes with very little effort.

[JF-AMD]

If I go from an Opteron 2435 D0 to an Opteron 2435 D1, my partners have to do a complete re-qual on the processor, that is how they work.

There are notifiable and non-notifiable changes. Doing a rework with HKMG would be a notifiable change to partners and that would force them into a requal. Requals take several months (complete reression testing on all platforms).

There are minor changes that we do that are non-notifiable (typically only touching one layer), but an all-layer change would force more work for partners. Most would not want to take the part based on the amount of work vs. the short life cycle. If I give them Magny Cours in Q1, then HKMG in say Q3 or Q4, with Bulldozer coming in 2011, they would either wait for HKMG (a bad option for us) or they would skip HKMG.

It's an economic decision, not a technical decision. You may see the change as low risk, but OEMs will not. There is too much money on the line.

[ajaidev]

JF-AMD i am a bit confused as to when documentation on sockets are sent to mobo manufactures, i know someone in a reputed mobo maker and he told me that they have yet to receive any documentation for the G34 socket but have all-ready received some info on LGA 1567.

Is it because AMD will have limited partners making G34 socket based mobo's or is it that the documentation is not released?

[mstp2009]

I can't speak for GF because I don't work for them, but think about the remoteness of the option to move away from AMD.

AMD is their largest customer. You make money in a fab business by keeping the fab running at 100% at all times. That allows you to depreciate the fab quicker and get on to the next process.

Who, exactly, would they replace AMD with? Who else in the business could replace that much business that quickly? If they moved away from AMD, they would have a very expensive fab sitting at low utilization and that would be very costly. GF and AMD have a vested interest in working together. All of the people thinking that somehow this deal is bad for AMD don't really understand how the financial aspects of running a fab work.

Fabs are most profitable when they are at 100% utilization. They have massively high fixed costs, and relatively low variable costs. So the quicker you can run through enough wafers to offset the fixed cost, the quicker you can get to the next node. By having AMD and other companies together GF can keep the fabs running around the clock, diversify their process technologies and increase their access to IP. All of that makes them more competitive.

Dropping customers, running low utilization and starting/stopping the lines will make the fab less profitable. There is a finite number of companies that you can strike a foundry deal with, so the last thing you want to do is lose one of the big dogs.

Not to nit-pick, b/c I agree with what you are saying, but the "who" that GF could sign would be ARM customers. ARM chips are like grains of sand compared to desktop CPUs. They are everywhere, in everything.

Sign a big enough ARM design (say Qualcomm and cell phone products), and they could potentially load up the fab with no need for AMD.


Just speaking hypothetically, don't flame me for it.