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.
paul In any case, it seems way too late to do anything but minor changes from either of the two layouts long since established. The MB makers would throw a hissy fit.
I agree. Although it is probably no big thing to change the socket designwise. But then you still have to evaluate the design before shipping out more than samples. If that has not been done already I see no way of Motherboard-availability in October. K.
paul - I think he'd had a few too many whenever he picked up this info
That is exactly what I thought as well reading it first.
But then, apart from "Socket 764" which is clearly a typo (or more probably the transcript of yesterday night's dull scribbles ;) ) those cryptic Pin-numbers could well make sense if e.g.
1. There are problems with Socket 754 detected. Yes, pretty late, but could be possible that this problem only occured recently with new steppings now in volume-production.
2. AMD learned more about Prescotts capabilities and concludes 128-bit memory channel is indeed needed for launch already to prevent from any QS-rants after Prescott-launch - unless anything above 2 GHz is achievable.
Boths ways it would imply some different pinout from Opteron 1xx - anything between 755 and 939 pincount - which would somewhat fit the cryptic numbers of the article.
K.
wmbw
"I think AMD has an exceptionally strong product with Opteron, and for the first time, they can actually stand on their own as an alternative platform. I just don't want to see their success tied to how well Intel does with their products. AMD should be their own entity and be in control of their own destiny. Just my $0.02."
Add my 2 cent for this.
As for the chicken and egg problem, a very easy one. Nobody would consider porting software if there is not hardware available for it.
As for software support for AMD-64, what I see in serverspace is very convincing: Operating system available at launch date, more coming every month. Including Windows. Databases, etc etc. I would say the present infrastructure less than three month after launch is nothing short of impressive. Which is only possible (without spending cash AMD cannot afford) because porting is cheap and easy.
For Desktop-segment, I like the the picture of the snowball turning into an avalange. Which implies there will be some patience required until seachange will happen.
Adressing the competitor directly, just a matter of taste imo.
I dont think AMD cares at all about Itanic - leave alone positioning Opteron against it: Which position would that be?
K.
wmbw
"AMD needs quite a lot of heavy lifting to overcome the barriers of even porting a software suite to their measly 64-bit extension."
Guess you are not talking about Server-stuff.
For desktop-suites: Sure. Strongly supported by the fact that very few software-suites are processor-starved these days. Although the technical barrier of porting to AMD-64 is fairly low.
However, there are applications that are hardware-starved. Grafics-applications, e.g. - Across the board from desktop-publishing to CAD. Speech-recognition, maybe, although i am not sure the bottleneck is more the juvenile stage of code here. Games.
Plus everything we know of or do not about is not yet implemented in software-suites for the current installed hardware just lacks to perform sufficiently.
Might take time, yes. But PC buying decisions are made on a horizon of four years. (Or, better said: should be made...)
From a pure rational base (assuming price-levels to be roughly equal) this is nothing short of a no-brainer. (Now, i concede decisions are far from being made from pure rational base.)
As for the strategic plan "to let the industry work for them": Everybody in this industry will continue do the work for his very own success. AMD-64 is just offering an opportunity to do that in a direction AMD will capitalize on as well.
In essence: Your DOF (Doom of failure) is just another word for FUD.
K.
yb
The time between two nodes is probably what you mean - or what was referred to in CC. Two years, roughly.
K.
yb - i missed that
And I doubt that it would be realistic to assume they could do that:
"65nm" is to be understood just a semantic reference for the shrink to come after the next, which is referred to as "90nm".
The latter will start end of this year (Intel), so probably "65nm" will be end of 2005 (Intel). AMD/IBM will be a year later. 2006. And I would not be surprised at all if it is going to be 2007.
All that if there is not a lot more hurdles in the way from the laws of physics and chemistry (for all in the industry) that we dont know yet - but have some indications for already, e.g. looking at Prescotts die and at its roadmap indicating poor scaling clockpeedwise within the next year. And listening to IBMs terminology: Interesting enough, IBM is still referring its next year's process (for Power5) as "130nm". Both are right in their references, Intel due to large amount of ultradense cache is sure closer to 90nm on average and can call it so, IBM just decided to be more conservative here, indicating current potential for shrinking the logic side is too marginal to rectify to call it the next node already.
The decision to migrate to SoI is probably an indication that such hurdles for shrinks below 100nm or so going forward have been anticipated for quite a while in Sunnyvale.
K.
DGagnon819
You would never hear public admission of a broken process from any company in this industry fixed by somebody else.
As for link, the amount you find in AMDs Annual Report 2002, which is a good read for other things as well.
As for IBM-AMD joint development:
IBM has a four-year experience lead using SoI. AMD uses IBMs SoI-Design Library resulting therof.
AMDs SoI-process is significantly different from IBMs currently. E.g. they use different Wafers, to start with.
As for what IBM is interested in is probably less the (limited)experience of AMDs SoI-process-technology but more the contribution of AMDs R&D-Ressources for future nodes.
FinFet-Transistors, things like that.
K.
CombJelly thanks
Although the article is fairly old, it could well be still accurate:
"Later, in mid-2003, the rest of Building 323, about 34,000 square feet, will be tooled up to become the development line that will continue to advance work using 300-millimeter wafers."
Considering the rest of the fab is not fully utilized already, it would make sense to leave the development line where it is now, delay tooling of the rest of the building until 2004 (if my model is anywhere near accuracy, AMD does not need new capacities before H2/2004, more probably only in 2005, depending on future SoI-Yields). 34000 sqfeet for modern automated 300mm tools, what would that be in Waferstarts for K8?
1000 per Week? Ideas anyone?
Now, if its that, it looks like an interim-solution until a new FAB (joint or not) will be going online for 65nm-node in the 2006/2007 timeframe.
Wbmws thoughts about the commercial organization of it would fit pretty well into that. As it would fit into AMDs efforts to increase the variable part of the cost structure.
For a while, that is.
K.
sgolds re: socket 754 has enough pins to support a second memory controller
No need for supporting a second memory controller. There is only one on K8.
What is often referred to as "Dual DDR" is in fact single channel, but 128bit-wide instead of 64bit.
Just in case you are speculating about specs for Athlon64-launch: 64bit Channel. Confirmed yesterday by AMD Germany.
Confirmed as well: Current K8 Memory Controller is not capable to support DDR-2......But implementation of a modified Memory Controller for DDR2 "technically trivial."
However, no clarification given whether Socket 754 allows 128-bit channel technically:
....change memory decision if market requests... no roadmap for if or when ... cannot confirm socket-specification for 90nm-successor.
K.
DGagnon819
In CC Q4/02. Although they did not elaborate for what exactly they paid for. IBM itself downplayed the consulting side of the payment. So it could be mostly licence-fees for SoI design-libraries AMD used since 2001 already iirc.
This by the way had nothing to do with the joint development agreement.
wbmw re: So you're saying that after all the investment in 64-bit compiler technology, AMD will finally match the performance of Intel's 32-bit compiler.
Apples versus oranges. I did not assume that "what I have seen" was referring to Intel legacy compilers. Irrelevant in today's HPC already - leave alone tomorrow's.
K.
wmbw
The optical impression of the chart is indeed that everything except galgel and mgrid looks mediocre. But overall improvements average 34 percent. Exactly what you claim it was behind Intel compiler before (from results that you have seen). For the world who has not seen the latter results the message is that SPECfp from next week going forward will indeed propel it significantly ahead of this week's Specs.
K.
avatar
In Semiconductor language, "future shortage" is an acronym for "current overcapacity". K.
sgolds
AMD did well not to communicate Opteron as an Itanium killer.
But considering K8 is still in its early juvenile stage clockspeedwise and considering there is a lot of headroom in FP to be realized from next week going forward I would be surprised if Opterons will not in fact be closer to Itanium than to Xeon performancewise at this year's end already.
As for "owning the computing world right now": Naah...
Its a lot more than financials alone. I dont see any evidence at all a K8 Desktop-product is manufacturable in volume (of some millions per quarter)- at least not at the current node.
And next node's product is still a year away from today.
K.
sgolds: Would be very surprising if not...
Here is why: http://www.investorshub.com/boards/read_msg.asp?message_id=1162606
K.
sgolds
Guess your conclusions next week would have been as significantly different as a SPEC-landscape of next week will look different to this week's
K.
sgolds
Guess your conclusions would have been significantly different from a SPEC-database of next week....
K.
wmbw
Would you care to take a second look?
(I did get the impression you mentioned as well on the first glance) K.
elmer - IBM must have a plan that considers the possibility of default
Sure. That is what I had in my mind with a "more sophisticated sale-and-leaseback-structure". In which case the parties would need to find a financepartner (better said: a consortium)accepting this risk and/or being able to find partial of full insurance for it.
Way too early to speculate about feasibility now. Lessee how the balance sheets look like after FASL consolidation, then lessee how 2004 is going. Hope what we see then is good enough already to give Bob a lil bit more than just one choice
K.
yb - If you are saying that AMD must be careful here - I completely agree
Yeah. Sort of explains Athlon64 to be released with 64-bit wide memory channel only as confirmed yesterday by AMD Germany.
K.
yb re: ... I post too much
Naah. Keep it up, man.
About the compiler-performance. Here is where Mike took it from.
http://www.pgroup.com/images/pg50vpg41.jpg
Now, kauftAMD ( http://www.wallstreet-online.de/ws/community/board/threadpages.php?fid=0&tid=629588&page=-1&.... )
was kind enough to educate me what it is all about:
Obviously the announcement of acml and preannouncement of SGI 5.0 not coincidentially came up, but correlate in many ways:
Both are in some way the results of the very same efforts.
The impressive performance-gains come from the combination of both using acml-code and the compiler-development.
I begin to get sort of an idea now Opterons reportedly hardly warmed the coolers up to now and why AMD was so reluctant in giving actual thermal figures: The CPUs seem to have been sort of underutilized
K.
yb - I'm sure IBM is getting something without the president face imprinted
Well, way too early for me to be sure of anything that will happen in 2005 - or not.
A simple lease of a fully equipped fab is a possibility as well. Or in a more sophisticated sale and leaseback - structure. Or one of many many other possibilities. Which ones will be feasible and which one will be the structure of choice can hardly be judged from today's viewpoint.
However, I agree in the core of your posting: IBM is confident in its development partner. Maybe even confident enough to leave the definitive agreement open until 2005.
K.
wbmw
Although what you assume the agreement could be is very different from a foundry-model, I would agree it is one of the best of many possibilities to play it - if IMB building a fab for AMD is accurate at all, that is.
Sort of a very creative synthesis of pros and cons of manufacturing versus outsourcing, very favourable for AMD.
As well a synthesis in respect of some of Jerry's most famous statements:
"Real men have fabs" - "We are very good in using other people's money"
K.
gollem It is true for mp3 players, cameras and pda's isn't it?
Affirmative.
gollem - Aren't phones supposed to be using Nand?
In the past - not that I would know of.
Presently, I dont know. At least one chipset i know of already allows it.
In the future, probably yes. Hard to say how many of them, depends how reliable Nand-based handsets will prove to be - or not.
K.
elmer
Maybe true. Could be they want to wait with it until DDR-400 ECC is JEDEC-specified and at least samples of 2GB-Modules of it are available for Benchmarks. And/or they have Models 246 ready.
K.
elmer
I see. Well, I think these guys are smart enough to know exactly what they do and what not - and what they mention in announcements and what not, would you agree?
K.
Elmer
http://www.investorshub.com/boards/read_msg.asp?message_id=1154973
Maybe you can comment?
As for Benchmarks, APPRO published Specs for a 4 way system. Three months ago. Could you help out with a link for Gallatin-SPEC-numbers @2,8? Or @2,5 at least?
Thanks
K.
That is one side
The easier and cheaper one. The other is Software. Which is the more difficult and more expensive side. Even more compelling.
K.
wbmw
What the headline of this paragraphs says to come.
Anything. "Is shipping", "available now". Whatever.
K.
Pricing and Availability
http://www.intel.com/pressroom/archive/releases/20030630comp.htm
See last paragraph.
Hmm...
Maybe they just forgot to mention.....
K.
sgolds
I see your point - and yes, maybe many AMD employees may take it like that.
However, I am deadsure there is already spreadsheets circling inside AMD to determine the quote of equilibrium for the day of the grant for every personal options-portfolio. Plus scenarios for future developments.
And a lot of time spent to discuss what to do within the next couple of weeks.
Well, at least something is offered. And as long as there is hope for personal gains, there will be no complaints.
Finally, probably I overestimated the impact of this program at all.
Options exchange program
Although I seem to be the only one attributung any importance to it, I went through it and find the following implications:
1. Strike for new options is final quote of some date not exactly specified, likely somewhere in February or March 2004.
The term creates the Paradoxon that rising quotes from successful operational results within the next seven (or eight) months will counteract financial interests of participating employees. (Top-Management is not egligible to take part in it)
2. There is a forecast for the AMD-quote implied to be around February 2004 at 12 to 13 Dollars. (exchange of lowest-strike option Black-Scholes-flat)
I am well aware that it is a very complex matter to design a program like this in a way of balancing interests under consideration of tons of legal, fiscal and accounting stuff, especially right now when handling of ESOPs is in controversal discussion. And I recognize the very best intentions the authors had with it.
But the procedural solution of choice I see here is not one of impressive excellence; carefully said.
K.
Haddock re: My thinking is that they would do better to have different names for different chips
That is what my gut says as well. Plus, it groans, it does not like the 256-KB product at all.
But methinks AMD is exactly right to to it like they intend to:
Halo worked for 386 (SX), it worked for MHz and it is very likely that people will continue desiring to buy a noble label with an affordable price-tag on it.
Now, introducing a 256KB-L2 Athlon64 could be considered as another paradigm-change at AMD from the Jerry to the Hector Era; i.e. stepping away from offering the best feasible price-performance ratio and educating the world about it to just meet customers demand.
First hand, that seems to conflict somewhat with Hectors "Value proposition" statements. In fact, in a more sophisticated view it does not at all. It just depends on the understanding of Value:
In a technical sense, a 256KB L2-CPU is nothing less than a step back (or two) and not really a value proposition at all.
But why that is so is nothing short of impossible to educate the very very vast majority of potential buyers of PCs, would you agree?
In a broader view, satisfaction is a value as well. So if you just satisfy people by fulfilling their desires it sure is a value proposition. No need to educate people as well, which AMD tried in many many attempts not only unsuccessfully, but worse than that, confused people with it. Confused people buy the biggest brand.
A propos biggest brand: As we know already, Intel is able to manufacture incredible dense cache. Guess what next year's benchmarks are skewed to. Guess what you will read in next years reviews about the model-ratings for 256-KB Cache-Athlon64s. Interesting to see how this will work out....
K.
ESO-Exchange Program is in force
Details here: http://amd.edgarpro.com/Results.asp?criteria=ticker_symbol&cols=7,0,4
Did not go through it yet, but last minute changes were made for terms (12 instead of 14 Dollars strike).
Next week seems to become interesting then.
K.
wbmw
Very stringent thoughts. Spinning it ahead, the claim Stephen made seems unlikely to be real. In case the vendor was HP a year ago, they sure would have offered a Madison replacement for free before anything else. And the customer would likely had accepted considering he is working on IA-64 software for a year already. In case Stephen is not with HP, as you said, it does not fit as well.
K.
OT sgolds
Thank you. However, the posting raised a question here which i asked myself today as well. So my posting was more putting my thoughts about it in a row. Doing this I use a piece of paper (or a posting screen) usually. Then, if I think it could be of any interest for others, I click on submit (even if more questions are raised than answered), if not, on cancel.
In essence, nothing emotional at all involved necessarily for the author of a posting that inspired me to think about it or coincidences with my own thoughts.
In short: I dont care at all who is posting, but what is contributed first hand.
Klaus
(Edit) Elmer
That would be the second version I offered: Backend Costs.
Which one to be used reasonably is dependent from what extent you believe throwing the chips on the market will cannibalize your other sales, what utilization you have plus some other considerations you know better than me.
That is why I offered both versions.
There is an even more extreme version as well: Before even thinking of revenue versus cost considerations, fact is AMD needs every Dollar for revenues it can get for Q2 to keep its miss in limits. So even giving it away for less than incremental costs of backend could make sense.
Klaus
p.s: Glad that you did bite the moderation bullet without any damage. Chapeau.
bg
Could be some disappointment about the fact that employee-stock-exchange program has not been announced to set in force already.
From yesterday's SC TO-C (here: http://amd.edgarpro.com/Results.asp?criteria=ticker_symbol&cols=7,0,4) it should indeed be in force already.
Maybe there was just a formal issue that had to be sorted out in last minute. If so, it signals that they feel urged to to it this week, otherwise they would have said so and postpone it until next week to sort out things without any rush.
Filing today, you would not be able to announce it during trade hours without hazzle. So expect the announcement after hours.
Could easily be a misperception of the market in other respect (which is admittedly a very gentle interpretation of AMDs latest actions thisrepectively):
They rang the bell twice now for investors with ears for it before taking action.
If so, the market seems to be too suspicious to believe in that. Or, almost nobody at all is tracking SEC-reports to take notice of what's going on with the matter.
Finally the conspiratious version: The boys playing the stock make sure there is nothing suspicous going on in the stock for three days preceding an important event moving the stock significantly. ;)
K.
yb - Do you mean they will stop selling anything under $50?
They are sure trying to do so, but short term I am afraid there is much chance to sell their lowest bins at or above full costs. Q4 would be the earliest when I could imagine this scenario.
As far as the rest ist concerned, I sure hope there is not much coming out of current production anymore in the 2000-2200+ range. But there is obviously still low-bin Bartons (not much I hope) plus Barton-dies with Cache-defects in stock which will come to market as Thortons soon (the first from earlier production lots of Bartons, the latter from all Bartons produced to date). In essence, I dont see downbinning at all with these products, but a decision to sell it at any price above their variable costs (maybe even at or above backend-costs) instead of throwing it away. Nothing wrong with it at all for me, especially if you can take some Celeron-shares with it and/or force Intel to give away Celerons to DELL even cheaper as it does now.
K.