Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
imho
I completely agree in what you post, mate.
All sources are subject to reality at best, and ulterior motives, at worse.
In particular in this. I am aware i got caught by the latter in the past more than once, and by the former very often.
I am always aware of both. But this does not prevent me to fall again for either of it, on occasion. All i can do about it is to make sure my learning curve is not hindered. It cannot be accelerated beyond this. You can't accelerate pregnancy working overhours.
K.
chipguy
Interesting. Snow has suggested so, as well.
http://www.investorshub.com/boards/read_msg.asp?message_id=16638146
See reply.
K.
alan
I think all the new features that are in Penryn (whatever they turn out to be) are in the existing conroe design, but not yet validated or functioning in current silicon. There are just not enough transistors available to have changed the design...
Hard to understand how features can be designed in without transistors. However, i agree in your understanding many features to be deployed in Penryn are in CMW-design already. Been told A0-Silicon of Penryn is best understood as being a revision of CMW at 65nm with 45nm cache attached. In so far i followed recent discussions of how fast current silicon can run soandsomuch Os's and apps with somewhat limited interest. :)
K.
Snow
Willamette was after Tualatin, not Cumine, as far as nitpicking is concerned. Irrelevant, naturally.
Seems to me it fits in the Willamette to Northwood analogy a little better than the P3 to P4 analogy.
Well, if any it would be more appropriate to compare it to Northwood --> Prescott. Actually, I agree. This analogy is more suitable than Tualatin -> Willamette in a technical way of looking at it. However, in a marketing way of looking at it, in terms of cycles, you know, the original analogy seems to be closer. I don't know. It's a bit of both. Depends on the angle you look at it what to prefer. Taking into account Nehalem is just a year after Penryn, i am slightly leaning towards to the Prescott-line of thought.
K.
smooth
Yeah, right, you changed the argument.
Not at all. I just expressed how i understand Mary's timeline statement. Wasn't the thread about it?
BTW, I think you might be in for a surprise on Nehelem. Are you expecting me-tooism in this situation too? Rather necessity. Intel cannot scale Chipsets further due to pad-limitations without integrating Memory-controller to Nehalem. It'll help performance. And 45nm should be ripe by then. The addition of both should result in a competitive product.
K.
smooth
Hah! I'd buy them for $9 too...
I'd not particularly surprised if there will be times when you can in the course of the year.
K.
mas
Parametrics. Static power and variability issues for SRAM. The latter expected to become better over time, the former not much. Neither design nor process team should be blamed for. "It's about what we expected". As for logic, verbatim: "Working on compromises as good as we can with the timeframe we have. We call it co-optimization. Sounds much better." Needless to say, no link.
K.
elmer
There is silicon. Ergo there is data.
K.
elmer
Poor in the sense that you form an opinion with very little or no data.
What makes you think i would ever form an opinion and get vocal about it just because i feel like, with very little data, let alone with no data? The part where i go for a limb is there is close to a year left data could change.
K.
elmer
Picking up dumb money from the street is not fraudulent afaik.
And yes, i think of Penryn in the same way i thought of Willamette: An inferior product to its predecessor. Again, it's neither fraudulent to offer inferior products nor to put wool on the eyes using a media-mindmachine making it shiny as polished brass.
I am well aware going out for a limb on Penryn, as no specs will emerge for quite a while. That is what i can be blamed for - and confidently expect this if Penryn will turn out to be a different animal than i currently see, i.e. i have connected the nodes wrong. But then, what is posting on investment-boards for if not going for a limb occasionally and watch what comes out of it? :)
K.
p.s.: When i recently mentioned a SI-ban is a mark of merit i did not expect you would be awarded the big mark. ;)
Smooth
So it will be 3-4 years before anyone uses 45nm.
Certainly not. There is always people buying stuff because it is new and Intels media-mindmachine says they need it because it runs benchmarks fast. Penryn is in time to reach those who fell for Netburst last time. So nothing much can go wrong as long as Intel has Nehalem in the fabs on a second gen process before public awareness rises these people have been caught again.
Nehalem is expected deliver benefits. But then, it will only be mainstream in 2010. That's what Eric sees, and Mary adressed with the 3-years remark. I see the same.
K.
Smooth
Huh? Where do they get this stuff?
While i don't happen to know Mary Olsson, her employer is not known to spread any nonsense. He was principal analyst at Gartner before he opened his own shop last year. Maybe just because he hates to put wool on public eyes ;>) Anyway, the chap not only knows a thing or two, but knows almost everybody in the industry. So it would be safe to assume Mary's remark has a credible source.
The remark certainly implies a suspicion 45nm could already turn out to be the inflection point where leading edge becomes bleeding edge.
K.
wbmw
when the figure was supposed to be 49 +/- 2%
Getting a tad closer now. However, not yet in line with what Andy says. Look again: http://www.intel.com/intel/finance/bus_outlook.htm
To make things worse, Andy changed wording from his usual "couple of points" in Q1 to "a few points" for the fiscal year. However, no reason for panic: I'd assume "A few" is limited to single digits in common sense.
K.
wbmw
talk about hitting the target but missing the point.
Yup. We are in the same camp here. That's what i expect as well. :)
K.
wbmw
Intel already guided towards GMs near 49% in the first half, plus or minus one percent.
Interesting.
http://www.intel.com/intel/finance/bus_outlook.htm
K.
WT
Most of it. Not all, though. Code using instructions not supported by AMD will be preferably used for benchmarking, if the world continues to turn the way it always did.
K.
Katie
Or the pun on density. :)
K.
sdwalks
all the multinationals were having wet dreams about entering the Chinese market.
And still have. Intel explores into moving waferfabs to China for a second life, and to India as well. Good luck with it....
I know China good enough i can confidently say chinese people want what the rest of the world has. You can't satisfy them with less. Even not if it's dirt cheap. Neither now, let alone going forward.
K.
wbmw
keeping Chartered around at the costs involved seems very perplexing.
Chartered costs are variable - although the cost occurs earlier than the revenue from it, so some of it is in different periods, i.e. quarters. It will take some time for the guild to model the seasonal patterning of this, apparently.
Chartered does not seem to be suggesting the same progress in yields as before
That is the picture i see as well - no indications this would change dramatically anytime soon. Maybe I find an opportunity for another crab-meal in Changi Village later this year.:)
K.
sdwalks
The Chinese clone of the plant produces cars for the internal Chinese market.
This happens. The clones are unreliable at best, and in some cases very very dangerous.
The same thing is already happening with various ICs.
True as well. However this does not work for X86-MPUs comparable to Intels and AMDs brands. Any market smaller than the planet is too small to make a business out of it.
K.
Chipguy
Thanks for the kitchen-picture, I love cooking. I agree all a development consortium can provide is a basic recipe. It is up to AMD to refine it and add flavour and spices. That's what they did for 90nm and brought a delicious meal to the table. Currently preparing 65nm.
K.
Chipguy
And just this week the third biggest semico in the world
has announced it can't afford to stay on the process
development treadmill past 45 nm.
Yep. Joining AMDs camp. Not sure you realize where this eventually leads to: Intel versus the rest of the world. Do you find this imagination pretty? I mean, for Intel?
K.
Sarmad
If AMD's retardation increases to 2 years (from the current 1.2 years), wouldn't that be a key disadvantage ? maybe fatal?
Rough seas ahead. Tough ride for AMD. As long as it continues to find supporters for its mission, they will survive the storm. Captain and crew are stormproved.
Yes, the question is when ? Is that next year (at 45 nm).
is it in 4 years (32 nm) ?
in 9 yrs (22 nm) ?
Interesting. Not sure you were aware what you scribbled down here: A very very long 32nm-node. Actually, that's about what i see as well. :)
K.
sdwalks
there are companies that will ignore IP and produce a competitive work alike
Who do you think of, specifically? Besides, i see basically nothing that would prevent anybody from producing it. However, selling it would be a problem.
The anti-trust cases against all players has removed the ability to protect IP.
I stumble to understand this part. Could you elaborate a little further, please?
K.
Sarmad
Are you saying none of the credit goes to process/node advantage ?
No. Neither I claim there is no returns left presently and going forward for the next two nodes at least. Diesize-economies still exist, it's the parametrical returns that diminish currently and will finally go away. Not the question of if, but when.
The law of diminishing returns was observed in agriculture by Gutenberg. So it is actually better understood as a natural than an economic law. Moore's law is basically nothing else. There is a linear part of it, which we have seen for decades now and there is an asymptotic part of it, which we already begin to see.
In a nutshell, there is times when you can extrapolate the past to the future. And there is times when you cannot. (More precisely, you always can. But things do not always develop like this.)
K.
smooth
AMD's 65nm process is lackluster in regaining the performance lead. Too early to tell.
Not to mention late to the party....
Returns of being ahead in nodes are diminishing. And Intels manufacturing strategy is at risk.
Think of: "So the last shall be first, and the first last" in the big frame of things.
K.
smooth
Apple is a good example there is no viable niche: They bailed out of CPU-design and use vanilla now.
K.
Vattila
nor is that a sustainable business strategy (despite analysts dismay at the current fighting). In simple terms, fight for share or die.
Yes. If things are simple, there is no need for sophistication.
K.
elmer
Just listened to the Chartered Call and learned they plan to expand their fab7 to 45k wspm versus previous plans of 30k wspm.
To be executed on after 65nm is ready next year. Customer demand, they say. (But not which Customer(s) :) )
K.
Katie,
see the post to wmbw pls. :)
K.
wbmw
Copper they got from Motorola, and implemented in year 2000 at the 180nm. Intel was a year behind when they implemented it at 130nm in 2001. Partially depleted SOI they got from IBM, implemented at 130nm, and is of highly debatable value. Intel has not yet chosen to implement any kind of silicon on insulator technology. Strained silicon they got from IBM, and implemented into 90nm by the end of 2004. Intel pioneered this technology in their 90nm process that started shipping end of year 2003.
Following you line of argument consequently ends up noting Intel did neither invented the transistor nor the use of silicon to fab fab chips.
Bottom line, i did not mean to suggest AMD invented the wheel, but that it makes nice vehicles out of it.
K.
smooth
AMD pulled Copper, SoI, strains in Dresden, cumulating in a process orbiting Intels in the 90nm node. Spells not only competence. Spells excellence.
K.
Smooth
Mentioned one reason in another posting you replied to. Another one is that it is hard if not impossible to get into agency business, military, administration etc. and into corporate business if you rely on foundries typically in asia. AMD could not touch these fields much - but it is essential to get there to stay in the race longerterm. AMD rightly said there is no viable niche-strategy in the X86-MPU-biz going forward. From this stance, the brutal-force growth strategy they currently pursue actually looks more like a necessity than a choice.
K.
elmer
Beats me. A portrayal of AMDs flex-capacity at Chartered not long ago (04/05-timeframe) indicated there is limitations. I'd expected it shown at capacity of the foundry-fab running at 110% utilization, 100% yields, and for current design fabbed in a node three years later if there were no limitations. ;)
K.
elmer
Yes. For this decade, at least. This was behind my remark the topic fits on an Intel thread.
However i am not sure this could change in the course of a decade. Looking at the the trend of sharing ressources of R&D, e.g. how the Consortium around IBM develops, it well could lead to downstream sharing as well over time.
However, for the time being process technology is a core competency in the MPU-business. If Hector gave it away AMD would never get back. Better to rely on subsidies for building fabs than to rely on foundry partners.
K.
Snow
Glad you like it.
Btw, good fun always has wisdom behind it.
http://www.nccbuscc.org/nab/bible/ecclesiastes/ecclesiastes3.htm
K.
DoU
Sounds right.
I'd wager terms of licence are a topic AMD wants on the agenda in negociations alongside the lawsuit again.
However i don't think Hector intends to go fabless. I am aware he was vocal about it many years ago as a possibility, but there is many good reasons that make this option questionable for AMD - at laest to a point where beggars cannot be choosers.
K.
Sarmad
The discussion was about AMD's possibility of going fab-less.
Oh. I see i am on an iNtel thread. This topic might fit there.
K.
Sarmad
It's 20% since 1994, when it was renegotiated in a settlement of a lawsuit, i've been told many years ago from a source very familiar with it. I take AMDs flex-capacity portrayals as confirming this number. However i don't look at it as a restriction. I am confident Hector has an option to buy Fab7 put in the foundry agreement with Chartered.
K.
Paul
Not sure you can trade up the shares you win in a competitive environment. What could be possible is to increase gm by means of lowering cost a tad over time. But even this is questionable in my book. So the benefit is really limited to building critical mass imo primarily.
Beyond, there is a side-effect to consider from AMDs strategy: It might disturb its competitors inventory management. I see AMD going for business in pricepoints Intel probably intended to ship P4. They still have more than one fab manufacturing it, so their inventory issues are likely to stay as long as AMD continues its path. Utilization of Intel will continue to suffer and make AMD look successful in competitive terms. Paul indicated Intel does not intend to aggressively defend business hurting its ASP in the best interest of shareholders. So what applies for AMD on the server side applies for Intel on this: They gotta live with it and concede unit shares in lower desktop and mobile pricepoints and bite the bullet of underutilization charges for a while longer.
Bottom line, it really looks like competition. No Nash-games.
K.