By the time the case clears all the appeals, AMD might no longer exist in its present form. As 3dfx found out, waiting for the possible settlement money to ride in on a white horse and save the day is not a viable strategy.
AMD needs money?
- Thread starter INKster
- Start date
- Status
By the time the case clears all the appeals, AMD might no longer exist in its present form. As 3dfx found out, waiting for the possible settlement money to ride in on a white horse and save the day is not a viable strategy.
It depends if there is a out of court settlement or not. If one does present itself, appeals would likely be factored out. Also, criminal charges might surface if Intel doesn't properly explain the lost emails, since it is now a crime to not properly back up emails.
This lawsuit is also being fought in Japan and I believe in Europe. So far the Japanese case against Intel is stronger than the one in the US.
I agree with that, but why would it be in Intel's best interests to do an out of court settlement? They want to make this last as long as possible, really...
And probably also as expensive as possible.
Im going to have to check on this, but I believe that Sarbanes-Oxley has a provision on email retention. If the court rules against Intel on the 17th it would likely bring the CEO into hot water. So, a settlement at this point would likely be in Intel's interest.
There is still a lot of questions in the air, it will be interesting to see where this lawsuit goes. The trial has already been scheduled for april 2009, so we have a while to see this play out.
epic
There is still a lot of questions in the air, it will be interesting to see where this lawsuit goes. The trial has already been scheduled for april 2009, so we have a while to see this play out.
epic
I've seen you mention this a couple of times, but couldn't find any obvious info about it. I'd be VERY interested in a source and some extra detail on that!
I'd have to admit that made me double check what I've read.
It's intermittently mentioned throughout the web, from Ars to the Inq.
http://news.com.com/2100-1040-275575.html
The above mentions this as well.
I don't know of a public version of the cross-licensing agreement, so I don't know what the exact wording is.
edit: This is the first I turned up.
http://www.secinfo.com/drdbh.a1v.htm
"6. AMD will have the right to use foundries for Am486(R) products
containing Intel microcode for up to 20 percent of its 486 production."
While this may be construed as not applying to later AMD x86 chips, the provision for Intel microcode may persist.
edit edit: There were later statements by Ruiz that make things kind of hazy.
http://findarticles.com/p/articles/mi_zdext/is_200201/ai_ziff21315
It either does not apply or 20% was enough to meet the objectives of that year.
Last edited by a moderator:
Thanks! That second edit with the quote from 2002 makes me think they indeed can't do more than 20% at foundries, but that they didn't want to do more than 20% there anyway that year.
That definitely sucks for them, and prevents them completely from going fabless. On the other hand, nothing in there prevents them from selling extra capacity to improve their economies of scale. But that'd require customers, surely! So that wouldn't really help here, and it'd put them in direct competition with Chartered which would be strange to say the least...
That definitely sucks for them, and prevents them completely from going fabless. On the other hand, nothing in there prevents them from selling extra capacity to improve their economies of scale. But that'd require customers, surely! So that wouldn't really help here, and it'd put them in direct competition with Chartered which would be strange to say the least...
It would also be a question of what customers there are that would want to run on the fab process treadmill with AMD.
Commentary on the x86 CPU world is that it's a very high-maintenance beast.
The processes are very focused on high-performance almost completely digital products. Analog and mixed-signal work may not be suited to the leakier and less hardy fab targets.
Process tweaks are common, and they are part of a feedback loop for core stepping tweaks and further process improvements.
Keeping a valid product on a shifting process would require additional work to make sure AMD's tweaking doesn't bork a previously functional design or tank its yields.
Outside of another high-performance x86 CPU manufacturer, the pool of customers that would want that kind of hassle doesn't strike me as being large.
They might want the technology, but the specific process is less compelling when AMD's emphasis lies elsewhere.
If underutilization is ever really a problem for AMD, I wonder if we'd see ATI GPUs on it first, just because those would have designers with the same direct fab access as the CPU group thanks to Fusion.
Commentary on the x86 CPU world is that it's a very high-maintenance beast.
The processes are very focused on high-performance almost completely digital products. Analog and mixed-signal work may not be suited to the leakier and less hardy fab targets.
Process tweaks are common, and they are part of a feedback loop for core stepping tweaks and further process improvements.
Keeping a valid product on a shifting process would require additional work to make sure AMD's tweaking doesn't bork a previously functional design or tank its yields.
Outside of another high-performance x86 CPU manufacturer, the pool of customers that would want that kind of hassle doesn't strike me as being large.
They might want the technology, but the specific process is less compelling when AMD's emphasis lies elsewhere.
If underutilization is ever really a problem for AMD, I wonder if we'd see ATI GPUs on it first, just because those would have designers with the same direct fab access as the CPU group thanks to Fusion.
I gotta think that AMD could get around this if it is even still an issue. They are no longer an Intel wannabe and have plenty of of IP and others things to use as leverage (ie anti-trust lawsuits). That is if Intel even cared at this point.
Unless you realize that your assumption thats it's more expensive to be fabless is likely wrong. I used the strikingly obvious NVIDIA example in a previous post. Here's another one - TI. They probably aren't abandoning fabs for the fun of it and they have the money to build them. WHy they didn't do it earlier is another question, but there is a lot of legacy there. NOt easy to make those decisions.
That said limbo is applicable to situation.
It's hardly a strong position in an anti-trust case for AMD to get a license to someone else's property and then distribute it for the rest of the world to manufacture.
The agreement that the 20% limit is tied to is the agreement that allows AMD to legally manufacture hardware-native x86 processors.
If that agreement lapses (I don't think it's forever, until 2009?) a lack of renewal or an end to the stipulation that the ISA is shared by all parties would mean an end to AMD.
If Intel is legally capable of adding an x86 extension that AMD is not permitted to copy (wonder about those vector extensions for graphics, out in 2009...?), then AMD is basically done.
Nvidia's volume in a tiny fraction of AMDs, or Intel's, who AMD would like nothing more to replace as the top CPU supplier.
The advantages of going fabless are:
1) You don't have to pay to build and operate fabs.
The disadvantages are:
1) You are competing with everyone else for floor space and are locked into contractual commitments, taking a great degree of flexibility out of your production schedule.
2) Since fab-farms carter to the most profitable common denominator, they are not necessary going to accommodate your design goals.
3) Fab-farms do not work for free - not only are they going to set the prices to recoup their process investment, but they want to make money as well. As such, AMD would have to price their CPUs 10-20% higher just to make up for the fab fees.
I think 65nm production illustrates this pretty well. I don't have hard numbers, but I would not be surprised if Intel has churned out 10 times more complex 65 nm designs than all fabrication contractors combined at this point. I would also not be surprised if they commence volume 45 nm production a full year ahead of TSMC, UMC and Chartered.
Well that kind of depends on how those graphics and GP[G]PU units are programmed at a high-level doesn't it? I mean NVIDIA and ATI seem to have done quite well in the graphics market without being x86-compatible due to the way their hardware is abstracted.
So for your scenario to pan out Intel has to add the extensions and ensure somehow that low-level binary compatibility is necessary or highly desirable. If x86 ceases to be where it's at then that presumably frees AMDs hand a fair bit?
Look at how well Transmeta did when it tried the same thing with x86.
Native support is pretty important.
If AMD doesn't have an x86 license, low-level binary compatibility might be illegal.
Either microcode or direct decoding of later extensions would count as low-level compatibility.
They could code-morph, for all the good that will do them.
If the equal sharing thanks to cross-licensing is taken out, but AMD can still maintain compatibility for prior extensions, then Intel can just add instructions.
It's corporate customers unwilling to chance crashing new software on their AMD systems and individual buyers hearing rumors of "broken" AMD chips that will end AMD as a CPU maker, since x86 compatible (can't legally say that anymore, either) products are its major revenue source.
For one, NVIDIA's volume is certainly not a tiny fraction of AMD's. In fact, if you think about it, NVIDIA just might use more silicon real estate than AMD. NVIDIA might be a fraction of Intel right now, but certainly not tiny, and if it's Intel MCPs capture a significant share, it might not even be a fraction in the not to distant future.
For two, you seem to be missing some fundamental economic logic that pertains to fabs. A) building the fab is an incredible cost and this cost must be applied (also know as depreciation) to future production. B) TSMC has a very low cost structure. One reason is that it operates in Taiwan and China. Cheap labor. Another is that it spreads its fab investments out over long periods of time. It doesnt just produce on the leading edge and retool, it sells the older capacity to many many other firms for years years later. Sure, Intel may produce chipsets at .13 or something, but TSMC is probably still cranking stuff out at .35.
But the very same logic applies to owning a fab yourself - you can use "older" capacity for other products in house. You have to deal with fixed costs and depreciation regardless of who operates the fab. All the factors associated with TSMC's costs will be included in the price they will charge AMD for production, including depreciation - as well as their profit on top of this.
The cost per-chip with in-house manufacturing is:
(Fixed cost of fab)/(# of chips over lifetime) + variable cots
Outside, it is going to be:
(Fixed cost of fab)/(# of chips over lifetime)* + variable cost + foundry profit
*The foundry still has to charge that in order to get ROI.
The time value of money is going to be accounted for in either cases - AMD will still pay for equipment depreciation, only it will be some else's equipment. You argument is that it's better to rent then to own. It is actually better to own then to rent - the problem is, you may not afford to.
When you also account for points 1 and 2 in my previous posts, I think it's pretty clear you should own your own fabs if you can afford to and know how to run them efficiently. Companies like Nvidia choose to forgo the financial advantages of owing fabs in long term due to enormous initial financial burden building one would impose. The reason that they can do so is because their competitors were subjected to the same cost structure. A febless AMD would be less competitive with Intel then it is now, due to production cost disadvantage this arrangement would impose
In terms of a car analogy, a fabless company leases a car (fab capacity at a given process), and after the lease expires leases another one. Company with fabs buys the car, uses it until something better comes along, buys something else while still using the original car until it has extracted the maximum utility from it, then sells it or breaks it down for parts. You will get a better ROI in the latter case, but it does require you to have enough money on hand to buy a new car every time a new model comes out. Intel does. AMD may soon not have.
Geeforcer - the world is not a vacuum. There are such things as cost advantages. These occur for a variety of reasons. One of them is economies of scale. TSMC has them. Then there are labor costs. The bulk of their employees are in Taiwan, including administrative and R&D functions.
Yes - TSMC has high a high ROE and high gross margins. But they have major cost advantages and are specialized. The proof is in the pudding. Look at last quarter for AMD - still producing at full capacity. Its average CPU price, even with pricing pressure, undoubtably higher than NVIDIA's average price. We don't know for sure, but given GPU die sizes, a decent chance that NVIDIA average die size is larger than AMD's. Yet AMD's margins, backing out ATI's, were lower than NVIDIA's. AMD is claiming the had some duplicate manufacturng costs, but as I mentioned in previous post, NVIDIA has been telegraohing it will raise it gross margin target. That puts them firmly up their with CPU players. How are they doing this buying from "expensive" fabs and with lower ASPs and large die sizes? Magic?
Yes - TSMC has high a high ROE and high gross margins. But they have major cost advantages and are specialized. The proof is in the pudding. Look at last quarter for AMD - still producing at full capacity. Its average CPU price, even with pricing pressure, undoubtably higher than NVIDIA's average price. We don't know for sure, but given GPU die sizes, a decent chance that NVIDIA average die size is larger than AMD's. Yet AMD's margins, backing out ATI's, were lower than NVIDIA's. AMD is claiming the had some duplicate manufacturng costs, but as I mentioned in previous post, NVIDIA has been telegraohing it will raise it gross margin target. That puts them firmly up their with CPU players. How are they doing this buying from "expensive" fabs and with lower ASPs and large die sizes? Magic?
Huh? Nvidia has a different business model which while quite lucrative at that, but has nothing to do with the fact that if they could afford to have a robust foundry operations they would make even more money. One of the reasons for Nvidia's high margins are missteps by competition, so a notion that their current margins are by themselves are any sort of vindication of fabless approach is IMO erroneous. Furthermore, the reason why CPU margins are rather low at the time is due to a price war. So no, it's not "magic", but rather a fairly obvious lack of competition in GPU sector and a heated competition on CPU sector.
Let us also not forget the tremendous amount of expertise that comes with operating you own fabs. You can taylor your process to you design as opposed to having to work in the parameters set by the fab, which are going to carter to the lowest common denominator. This is just one of several advantages to owning fabs that I mentioned previously without it being addressed.
Let us also not forget the tremendous amount of expertise that comes with operating you own fabs. You can taylor your process to you design as opposed to having to work in the parameters set by the fab, which are going to carter to the lowest common denominator. This is just one of several advantages to owning fabs that I mentioned previously without it being addressed.
- Status
Similar threads
- Locked
