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.
Kate
I think Intel will go for the gusto :) and keep it going on and on and on.
My pretty, why not look at it like taking away a toy from the big boy and give it to the small one who claims the big boy has taken it away from him while you didn't watch? :)
K.
justaview
The last tranche of this facility due to draw after reaching the milestone of 65-nm CPU-output. (i.e. launching 65nm product.)
K.
Elmer
Anecdotically, I have been involved in a U.S. ligitation some ten years ago. I can assure you, whether you are aware of any wrongdoing or not is meaningless for the feeling whenever you immigrate to the U.S. in these times. You know the proverb of being in court is as being on high seas?
K.
Duke
While Intel did not object as AMD said in the first paragraph of its release (from the context of the following paragraphs) the ruling is final now.
Kind of nasty for Intel. Its partners abroad won't like to get involved in this ligitation. They will put pressure on Intel to settle the case. Which is what I guess the judges intended with the ruling. Gonna take a ten digit figure now for a settlement, i wager.
K.
wbmw
More bad news for AMD...?
The platitude a silicon-shift takes longer than a socket-shift
blown up to an article? :)
K.
wbmw
I'm talking silicon with my picture, not software tuning: The parts to integrate two Brisbanes to Barcelona, and connect four cores to shared L3-cache have not evolved in Brisbane, but i believe the changes from K8 to Brisbane are intended for it.
K.
wbmw
That's not a problem with the arch, Klaus. For me it is.
When I look at an arch i ask not only if it makes sense, but also if it can be done, and if yes how long it will take. In the latter I tend to be too optimistic anyway.
..it does have some major new changes that will add risk.
Yup. Added on top of a geometry-transitions and a 300mm-transition.
What would you have done differently? Ask them to do more and hit a faster schedule?
AMDs schedule was as tight as it could be. I'd probably done something opportunistic. An MPC-Quad of 90 silicon by the time when Intel lauched Core-2, e.g. Nothing spectactular, just for an opportunistic bridge, allowing for a fallback to continue it with 65nm silicon if necessary. But then, K8 alone might have enough carry to allow this, if all goes well. I wish them luck it does.
K.
Snow
Certainly Vista will run on Core-2. What I compressed into shorthand is I have seen AMD working as close with MS over the last few years as MS released a development-vehicle to OEMs dubbed Windows-64. So it is probably fair to assume AMD has optimized its new arch for Vista. Its tapeout was closer to Vista-RTM as Core-2's, as well. The statement wrt performance crown made carefully enough not to say it necessarily will regain it. This depends on how good it goes at AMD and how much time Intel has to make revision changes until Barcelona will be in the market as well. With respect to the latter, a launch scheduled less than one year after tapeout for such a complex design looks ambitious.
Apologies the compressed portrayal has caused an itching at your head you needed to scratch. Hope it's gone away now.
K.
wbmw
Do you really think that Brisbane has all these latent powerhouse features just waiting for a new firmware revision to awaken??
Nope. I believe the benefits will only materialize after the rest of the organs of the fetus will have been evolved.
Nevertheless Brisbane should look better with its RAM properly clocked. :)
K.
wbmw
In a nutshell, I believe it is in the dimension of time-to-market concerns.
K.
Elmer
Q1/07.
Not sure if the search in ihub retrieves postings of 04 nowadays. I believe to remember it was limited to the postings for the current year, or a year back when I did it last year.
Not sure when i made this call at first as well. It must have been after AMD announced Fab 36.
K.
Elmer
Do you remember my call two years ago when I expect AMDs 65nm-volume? :) (It's on record on IHub and SI)
K.
Sarmad
When I said it is too early for conclusions, this applies for me as well. So below is just opinion.
Intel product is ready.
I don't like Barcelona. Without rationalizing it. It is just my impression from looking at the arch.
Intel leads the benchmarks currently. I can well imagine Barcelona will regain the Halo: Barcelona is a Vista-Product - and the benchmarks will be made on Vista. Then AMD will lead again. But I will still not like its arch. I could not care less than about benchmarks. And Intel probably won't care much as well. They want to make money. For their cashcows (Server, Business Clients and business notebooks) Vista is no issue for 07/08. Consumerspace, yes.
So, benchmarks aside, it depends which OS I run to say which architecture I prefer. And I never use Microsofts products before SP1. Maybe even SP2 for Vista.
K.
Sarmad
Maybe a tad early for such conclusions. There is a lot to consider.
First, the product introduced is at a development-point which AMD would probably not have launched as a product were it not a milestone in financial terms as well, i.e. to draw further tranches of loan-facilites. From this stance, its ramp might not be exactly as steep as many expect, if you know what i mean. ;)
Second, K8 until revision F was a great design for DDR-1, but not so great for DDR-2. So we can expect AMD has made changes in the memory system on revision G. However to make use of these it usually needs updated firmware and middleware (bios, drivers and stuff). When they shipped out CPUs for Previews (all saying preview for a product officially launched is telling) AMD could be pretty sure the reviewers would screw it up. Memory not running at full speed, timings not optimized, all this sort of stuff. And they could count on the habit of these chaps running their benchmark-tools (not even revised for the product) like monkeys and quickly post its readouts and draw conclusions from it without even checking basic settings of their systems. There are a few who would, though. Dimitri Besedjin, or Michael Schütte, e.g. AMD apparently did not ship CPUs to these.
Third, from what I see AMD does virtually nothing to promote its 65nm-product - which I understand as they want folks out there continue to buy 90nm. Bottom line, I believe the chaps at AMD know what they are doing and why.
So far I think it's fine and entertaining: The usual amplifiers of news go ballistic, everybody else can decide al gusto to add his speculation ad rem or ad personam comments - or to get couple good laughs from watching the scene.
If there is any conclusion you can draw from Brisbane it is for the current state of affairs of AMDs 65nm design and technology. On another thread i used the analogy of Brisbane as an ultrasound image of AMDs 65nm baby at its current stage. I do not see any indication for a down-syndrom or anything other bad. Some organs look big relative to others, some have not evolved at all yet. Nothing to worry about, but just as things develop usually. The fetus already got heartbeat. I do not hear anything suspicious, as well.
K.
Merry Xmas everybody. Klaus <eom>
Duke
This is of course, from the Glen Falls PostStar. :))
Maybe just a clip-paste-edit piece from here:
http://blogs.timesunion.com/business/?p=607
But then, from all arguments AMD ever made, Jerry's "we are really good at using other people's money" is probably the hardest to dispute.
K.
Elmer
Political correctness and parental authority don't ever work on adults
I am not sure this is true. The only thing in it i am reasonably sure about is it does not apply if you define adultness by age. I'd rather suggest to use an industry term: Mature. But even then I would be rather inclined to disagree. But this can easily be due to an alien feeling for english semantics only.
K.
drjohn
If I understand it right, mods at SI have the power for bans, but not for removing posts.
I believe it is not exactly a good system. Because it is a bad idea to make use of this power: From all (my) experience the more people get banned the more people will do their very best to pervert the system. People just behave so. So the most probable outcome of banning people (for whatever reason) is a fight against an increasing number of windmills.
Ultimately, a board organizing itself is the best, if not the only way to do it: If it is not capable of doing so it is moribund for good. That's how nature works, in my understanding. Any belief to have a better system was proven wrong so far.
K.
Edit: There is nothing wrong with moderation. A board organizing itself can have a moderator. But it is not necessary his alias is on top of the board. It should be one of natural authority. comp.arch works like that since ages. Nick very rarely does anything in terms of moderation. It is enough he is there.
Elmer
A little trip down memory lane...
How did it end back then?
K.
p.s: You really get a bad treat. Take it easy, mate. I lean more and more towards considering a ban over there as a mark of honour, if that helps. :)
Elmer
I invested in AMD with the understanding that AMD would field competitive products. I am entitled to compensation.
Be careful with this reasoning, buddy. Courts can issue committal orders, as well. :)
K.
drjohn
65nm with its high latency- lesser cache is prime class action lawsuit bait.
Only who bought such a part could sue, no? Go find one... :)
K.
Elmer
You now agree with me and Tenchu.
Not in all, as i said.
Are you feeling alright?
Chipper, mate. Btw i have flown over a pile of papers on dfm approaches suggested this year today. I apparently did not miss much. A lot of hype there, nowadays. And downsides of the approaches which looked promising yesterday are showing up. I need to talk to an old buddy at Mentor to get a better picture.
K.
Tenchu
Thanks. Thought so. In the tenthousands in units per year?
K.
p.s: From your posts i looked at recently, i seem to agree in everything - except for the conclusion a single supplier would serve the world better. However I understand the stance from an investment-standpoint. :)
wbmw
Is this top-enthusiust market relevant in any way except for the halo?
How many of these would you guess you sell in a year in terms of a decimal power dimension in units? (AMD FX and Intel EE)?
K.
wmbw
He notes, as well, that “a few notebook wins” that AMD thought it had “now seem to be in danger” and may be lost to Intel.
Not particularly surprising. Every design based on ATIs 690 parts is in danger. Not only the few notebok designs among these.
http://www.dailytech.com/article.aspx?newsid=5435
K.
wmbw
You assume this is intentional?
Yes. I have seen it too often now in the media which I know would not print or run it without Intels approval.
On some second thoughts of it, apart from the fact the species building their computers from retails components is dieing out, these chaps know anyway they can buy the parts cheaper in half a year. So the market impact is very close to irrelevant.
But this does not give the answer why they do it. It might be along this: The message transported to the public is Intel will be aggressive on prices next year. I believe they do it to prepare the financial markets for showing lower margins next year and at the same time insinuate a reasoning for it.
K.
golfburn
Oh yes. Channels have been informed. What is new (for me) is the public is informed, as well.
K.
p.s: Pls consider i did not follow for about a year. Pick me up at end 05. ;)
wbmw
These details seem early, but for a processor that's priced at $163 in Q1 and $113 in Q2,
Now that is list pricing, which I am well aware is different from prices OEMs pay. What I am somewhat puzzled about is why Intel nowadays communicates waterfalls to come. I am sure it is for good reasons. But I do not have the underlying theory in my back pack. Anybody has a pointer, or a concept for this?
K.
Elmer
The motto is never throw away a packaged part for any reason other than assembly related defects.
Now that's a concept as easy as even i understand it. :)
K.
Elmer
I see what you mean by "screening out" now.
Yield would then reflect only saleable parts
I give you couple examples to point out where the difficulties are in the terms yields and saleable parts.
The first die is one that works perfectly but only at frequencies that do not justify packaging currently. Frequency of your choice. But in a slower quarter, for third world market to keep utilization of packaging up, it would be justified. Saleable?
The second is one with excellent characteristics in terms of power, frequency and all. But you can't use the cache. There is markets for it, different packaging and all, but not as a CPU. Do you count it in for your concept of yield?
The third one works, but exceeds powerdraw for all your current specs of products. But this changes, as you know. You can change it yourself whenever you like. Saleable?
I could continue for an hour with the list. But you see where am I am getting at already, I guess.
K.
Elmer
I have no desire to attack or destroy you.
Well, I did not invite anybody to attack me - let alone destroy me. You read me long enough what is to expect when this happens. :) What I want to be attacked are the concepts i put on the table.
As for variation, I could not care less who says what on which board about it. As for sreening it, yes you can, but you cannot screen it out. Nobody can.
how big a problem it is apparently varies from company to company
I could not agree more. It depends in how far designs for manufacturability are taking the phenomenon into account. From what I see Intel seems to handle the issue better than AMD. Not on the manufacturing side of the house. It's in the dfm. No clue how they do it, though. They talk a lot. But not about this.
K.
Sarmad
..the burden of $4B in debt for an asset that is not only not paying its way, but is losing maybe $100m / year.
Maybe you are too low in the latter. I definitely agree in the burden. However I guess AMD had no other reasonable option.
K.
Elmer
Others may not agree with you.
That is what I hope for. I'm here to learn, not to make friends by sharing common errors of a BB. Everybody is invited to attack and destroy what I put on the table. I will do my best to defend mine, but i am willing and I believe capable as well to adopt other's concepts.
DD need not refer strictly to physical defects, but the number used in modeling to predict yield based on data gathered from some high volume process monitor.
Here it begins. There is good reasons for every of many ways to model. I personally prefer Bose-Einstein because it allows approximations for different design-complexities. Others might disagree.
It should include all failure modes which are screened at sort. Based on that, I have referred to.25d/cm2 as a baseline for world class.
It really gets beyond failures nowadays. E.g. you can have all transistors and all interconnects working (no failures) but still cannot use the die because variation does not allow it.
I'm not saying Intel or anyone else actually achieves this but if they claim to then this should be the measure.
I'd guess almost everybody has dd at or even well below world-class levels. SoI a tad higher than bulk, imminent to the technology's large defect areas you can not avoid.
What is your opinion?
The picture of yields of today has many shades and colors beyond couple of numbers maybe sufficient to determine yields ten years ago. Even back then it only allowed for a rough estimate.
K.
Elmer
How would you define "Excellent Yield"? I have defined it as a defect density of .25d/cm2.
DD is just one part of the yield concept nowadays. It is still important, and in this respect Fab30 is world class. This is not inside knowledge, but published, btw. However, many other things go into the equation. Referring to yields in percentages nowadays does not give a good picture anymore. Feature-yields i a more complete concept, e.g.
K.
Sarmad
See my reply to wmbw pls.
K.
wbmw
So you think the 45nm die will be intrinsically more expensive than the 65nm die? Why?
No. 45nm will eventually yield lower cost per die for the same design. Just not in its first months of production. And 08 will be early production for 45nm, while 65nm can be expected to be very very mature by this time.
K.
Sarmad
Can you give any detail or insight on why that is ? Is it scale of production ? Is it the fact of Fab 30 being on 200mm 90 nm for another 9 months ?
For the first part, Design for manufacturing, SoI, just to name two.
I am not getting at economies of scale with it. Just silicon cost per sellable die.
AMDs fab 30 is having excellent yield. No disadvantage from there. And it will be online for longer than what you assume.
What about at 300mm 65 nm, would AMD still have a cost dis-advantage ?
Yes. The two above still apply.
Would you expect a different yield between the two companies for 300mm 65 nm ? And do you have a feel for which one has the better yield ?
Yes i would. Intel is ahead in the yield-learning curve. And it makes a less integrated design (no IMC). And it does not have yield loss of SoI.
Guess you have the feel yourself now and do not need mine. :)
K.
wbmw
Cost is not unrelated to diesize - but diesize is just one of many factors in the cost equation.
Anecdotically, I had a good laugh today, being confronted with Intels 965 die being 130 sqmm or so made sb conclude it obviously would have the same cost as an Athlon die. I believe this example should make clear what i am getting at.
Two Wolfdale die (at <100mm^2) in an MCM are going to be far, FAR cheaper than a monolithic Barcelona core at ~275mm^2. And even package costs aren't going to bring Intel anywhere near these costs.
Possibly. Or possibly not. See below.
Of course, we are talking about 45nm Intel die vs. 65nm AMD die, but that's simply how things will compare in H1 2008, and I'm guessing H2 2008 as well.
Halfway. Yes it probably will compare as you say. No for a 45nm product being cheaper to make than Intels 65nm product by then for the same design. Counterintuitively.
K.
Elmer
Well, I believe any discussion of design and manufacturing reasons would unavoidably produce loops of postings as it always does.
So i prefer the easiest answer: Because it is Intels paradigm to deliver what it thinks it takes in the cheapest possible way.
K.
tx for explaining "mismay".
Sarmad
Now what we don't know is how production costs run for the two companies.
Mabe not exactly - but I believe there is pretty predictable trends: Netburst Silicon is way cheaper to make than Core 2.
Nevertheless, AMD has signigicantly higher cost of silicon. The gap could close if Intel would continue to make MCP-Quads in 45nm, which is quite expensive. For this node, I believe it's the better idea, if only because it's couple quarters faster to get to quad-products.
K.