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.
Mike
Well. A64 price increases and inventories go like horse and carriage here. But I'm not sure either. Backorder volume does not fit into the picture - unless these are figures AMD confirmed to deliver short term on orders.
K.
p.s: Thanks again for Level 2 data.
i-banker
Several independent reports of thermal issues with Prescott reported. But most probably not associated to cache, rather subthreshold leakage of logic transistors. Publications from Tokyo predicted this would likely happen already in 2002.
K.
i_banker
P4EE is made from Gallatin process (130nm). It is getting hot, but has apparently enough diesize to prevent from throttling issues.
K.
buggi
I did not even know S&P has a rating (apart from credit rating) on AMD. They are not listed as covering analysts, at least.
http://www.amd.com/us-en/Corporate/InvestorRelations/0,,51_306_657,00.html
Tewari. Anybody heard this name before?
Btw a while ago I kindly asked the credit analyst to elaborate on his estimate for option expenses.
His only reaction was to make acquisition people of the firm to spam my mailbox.
Bottomline: Standard. And poor.
Next topic, please.
K.
mmoy
Sure. I am one of them. And I might add turbine-characteristic noise makes me rather sleepy than excited. (which is a real blessing on long-hauls).
I've just thrown in these thoughts because I can definitely say it would be fairly easy to find quieter gear to do the job in server-racks for given footprint, airflow, pressure and redundancy. With some degrees of freedom of layout for a specialiced engineer, a lot more would be achievable in terms of noise-levels. Under these premises, low noise just does not appear to be a design target in my book.
K.
dan
I could well imagine server vendors deliberately choose fans generating turbine-like sound characteristics for the reason of performance associated with aircraft engines in our intuition.
K.
p.s: Btw, from what I read in notebook reviews and reports from folks, it sure makes more sense to focus rather on sound characteristics than on absolute sound levels for notebook designs as well. Probably the same underlying intuitional phaenomenon.
ddb
Could be. In this case temps would sound realistic, but the difference of values would still be questionable.
Couple of years ago I went for lunch in Taipeh with Sonja, a charmaing engineer working for a Mobo-manufacturer with this adorable kind of permanent smile on her face which always has a refreshing effect for me when arriving from Europe for a while.
Back to topic, she explained a lil bit about temperature monitoring, saying that due to the purpose of the circuits these are really only usable for the threshold-values where the logic should take any action, like controlling fans, shutoff, among others. All other values are just unsupported. Even at threshold values there is a wide margin of error, she said. I dont remember if she said plus or minus ten degree or within a ten degrees range.
But then, people want gauges. And there is binaries to make gauges move.
So, the values shown could really mean anything. Board-Sensors, CPU-sensors wrongly decoded, whatever you could think of.
K.
ddb
While I take temperature-displays of monitoring-software with a pinch of salt, the ceteris paribus comparison could be meaningful - although most probably not on any scale. However, i am not yet sure what to do with it: I mean, both readings of 46 and 61 degrees Celsius are shown as CPU-Temperature which is nothing less than just nonsense.
K.
ddb edit
Maybe I should have read your reply before answering paul.
Yes, completely d'accord. Max die temp was my bad, not Pauls.
K.
On second thought, thanks for mentioning the diesize-aspect of case temperature which I forgot to consider. Assuming Winchester and Newcastle on the very same powerdraw one would expect higher temperature measured at a spot on the case above the center of the Winchester-die, would you agree?
I dont have definition of Case-Temp handy, maybe I find time to look it up later.
paul - edit
It's max Case temperature, Paul, not max die temperature. But you are certainly right, I should have said "current specs". And, again, all this under the assumption HKEPC got it right - the K in the CPU designator-string is not officially linked to 60 degree Celsius Case temperature yet as we deal with a production sample of an unannounced product. I would not completely rule out it would show K as 70 degree instead for max case temperature in the docs to be released.
K.
Edit: Just saw max die temp was my bad. K.
Yuri
Well, the design of the heatsink is probably not a source of error as they apparently used the same one for Winchester and Newcastle.
On temperatures: you would excect die-temperatures in the eighties (or nineties) Celsius when running CPUBurn. Btw there is thermal diodes on the die which deliver values to ACPI. There is software available to display these in Celsius.
Anyway, there is one piece of information in the article which points to higher power consumption of Winchester-Cores indeed: (if hkepc got it correct, that is): Specification of max Die-temperature is 5 degrees lower than on Newcastle.
K.
Yuri
I would also say that a temperature of 65 C seems close to a die temperature
How would you get even close to the die with a probe of a digital thermometer?
K.
Yuri,
welcome onboard, to begin with.
On temperatures, my mandarin is unfortunately weak. So I cannot tell where exactly hkepc-folks have taken temperatures.
From the picture showing a digital thermometer it seems it is somewhere at the case level - not to be confused with die-temperature.
Whereever they have taken temperatures, if they used the same spot for Winchester and Newcastle (which I would assume) I just take notice they measured higher temperature for Winchester than for Newcastle period
However, waaayyy too early to draw any conclusions from that. CPU is not launched yet, let alone BIOS-revisions to support it. Somebody already mentioned CPUID shows voltage different from spec.
Suggest we give it a rest for a fourtnight. Reviews of a released CPU on platforms with released BIOS-updates will give us a much better picture than premature trials.
K.
deleted - obsolete
upndown
You might have a second look to the comparison you posted.
Btw, better cut the Spindler presentation out of the calendar.
I'm right now listening to it - it's actually a waste of bandwidth.
K.
upndown
September 23, 2004 9.30 PT
Frank Spindler, Vice President of the Corporate Technology Group and Director of the Industry Technology Programs, presents at the Banc of America Securities Conference.
http://www.veracast.com/webcasts/bas/34th-annual-2004/id13404262.cfm
K.
SC Online: HPC users are rapidly adopting the Opteron processor & PathScale compilers
Subject of this week's newsletter from supercomputingonline.com
Excerpt:
DOD, DOE Labs, Civilian Agencies, Life Sciences and Research
Universities Throughout the World Among Initial PathScale Users --
PathScale, developer of the world’s fastest compilers for AMD Opteron
processor-based Linux clusters, is fast becoming the number one
compiler choice for prominent researchers, scientists and engineers
who want to achieve the highest performance from their HPC systems.
Many of the highest-profile R&D organizations in North America, Europe
and Asia have adopted AMD Opteron processors as their HPC platform and
have purchased the PathScale EKO Compiler Suite to maximize 64-bit
Linux application performance. An important milestone has now been
reached with over 1,000 sites downloading the PathScale EKO Compiler
Suite.
More on http://www.supercomputingonline.com
K.
Close, Pete
If i got it right, the payments are for the period from now to to end of 2008. Which makes up for 17 quarters in my calendar.
Which could eventually be stretched, as payments due are bound on development milestones.
Actually, there is really nothing much changed from the existing agreement since end of 2002, it's just extended. The new one is a little bit more clear, as the last one had some rather strange things in it from an accounting viewpoint imo. Iirc this has been discussed here a while ago, i_banker was the one who brought it up if memory serves.
K.
wbmw
..IBM still has about $250-280 million worth of IP per quarter that they are putting into the project greater than what AMD can offer
An adventurous take.
The payments have really nothing to do with IP imo. I look at it as basically just AMDs contribution to the costs of running a development fab with all bells and whistles. The amount is fairly small because processing of development-wafers (and analyzing, etc) is done within IBMs foundry currently.
K.
yb
First, the analyst covering Lenovo is not Glen Yeung who covers AMD.
I believe Glen sees very well what is going on in China. If it does not encourage him to up his AMD-eps forecast it is either that he does not see any margins in the deliveries to China and/or his assessment is that retail pricing pressure from "re-export " activities will likely damage AMDs retail-ASP in other markets to compensate for any positive contributions from deliveries to China.
Which is not an unreasonable standpoint at all imo.
K.
Mike,
thanks for the data again. Good to see Semprons selling well and holding prices. At least for now, it seems AMD has managed to extend the tight grip of channels it always had for AMD64 on Semprons as well. (But well, the quarter is not over yet.)
K.
wbmw
Well, smarty, guess you need to review the link above. Opteron has 6% share in the Top500 list. Even Itanium 2 has 12.2% share. Care to amend your previous statement?
Not until supercomputer architects begin to choose processors based on marketshare of legacy-CPUs of the same brand.
Those i happened to meet are looking forward. Dual-Core-drop-in, you know. And by the way, nobody of them cares about Linpack.
K.
wbmw
Why not look up the number of Top500 supercomputers made up of commodity Xeon systems.
Can't find even one single system made up of EM64T-Xeons in the TOP500 list. Quite a bunch of Opteron-Systems in there, though.
Guess folks building supercomputers know a thing or two.
K.
wbmw
The performance boosts don't look like something that could be "faked" or "emulated", as the conspiracy theorists like to call it.
Certainly not. However, you might want to have a look at what Intel folks think about relevance of Linpack. (IIrc there is at least one or two presentations from IDF available for download). And that is for HPC-applications. For typical Xeon-workloads, Linpack performance means close to nothing at all. You should know.
K.
Keith
I wonder for how many more years they plan on artificially ignoring A64 for their Gateway lines?
Short term, its just an equation of how much is offered for doing so and the progression-rate of business-atrophy.
Longer term, the fact viable business-models exist for Baby-Bells does not necessarily mean there is any for Baby-Dells.
K.
Today the Common seemed to be astonished to realize having wings.
Wonder how long it takes to become aware of being an eagle.
K.
Many thanks again, Mike
Very, very interesting data. These "Level-2" you provide weekly help a lot to understand what is going on.
Reappearence of 1,8 Celerons is indeed wierd. Could it be a lot from Taiwan spot market?
K.
p.s: Hope you enjoyed your short vacation.
dacaw many thanks oem.
Any reports of how Labor Day retail sales went?
TIA
K.
mas
Well, a million Prescotts a week and a million Pentium 4´s a month. I see nothing really conflicting in the two statements
K.
mas
Here's another one from this interview:
Q: Does Intel see any advantages to its own implementation of X86-64 (EMT64 if I am not mistaken)?
A: At this point, we are working with Microsoft to minimize compatibility issues out in the software community before they go gold. That is priority 1, 2, and 3 at the moment.
Ok. Compatibility-issues cannot be resolved, just minimized, which is expected to take a while. Not a problem, as Win64 is still as far away as it takes from conception to birth for humanoids.
Anyway, it is some navigational aid of where EMT64 currently is: At least a year behind - if you close at one eye and let Intel get away with couple of workarounds.
K.
greg
Guess it really goes beyond margins. When I listened to Andy's midquarter-update I thought (among other things which political correctness would not allow to post publicly) "lots of folks in the food chain apparently are holding breath currently, seeing increasing inventories could make an opportunity to buy chips really cheap lateron for holiday season."
Which ultimately is nothing less than a trial of strength Intel is challenged to. Now, raising prices is certainly an option to respond to it if you can hold your breath as long as Intel definitely can.
For the time being, it is just intended indescretions of such intentions via the usual channels imo.
K.
Combjelly
Guess the speculation was dies are indeed side by side covered under a plate.
However, I dont see a good reason for AMD to hide an MCM if it were one.
Am i missing something?
K.
Shelton isn't that bad. Being a Dothan-derivative, it might not take a very big hit at all because of the lack of L2 cache. I am not sure it was worth the design effort to remove the L2, but that is another issue
Its Banias-die, not Dothan-derivative. Design-effort is limited to packaging. Probably you missed some pages of the article linked previously, I should have amended "ff". (You dont need to master mandarin, its in the pictures.)
K.
CombJelly
Quite frankly, AMD should put these in a uBGA package
I guess AMD would do so if sb comes along with a volume design in mind an asks for it.
As for the capabilities of Shelton vs. Geodes, frankly I have not much clue about the embedded space, but NX is a fully fledged AthlonXP CPU and Shelton ist really just the skeleton of a processor architecture. Guess it depends on the application what it compares with.
K.
Combjelly
I doubt Intel would get significantly more than its packaging cost. Its a capacity-filler.
As far as AMD is concerned, Geode GX would be the closest product to it, do you agree? How much do they sell it for, including chipset?
K.
neye eve
Think of .net as common interface between administrative and manufacturing s/w. Now, you obviously need software at both ends supporting it.
K.
Banias goes embedded
http://www.hkepc.com/hwdb/intelobc-1.htm
While being a possibility to fill some packaging fabs with it and corresponding chipsets, it is certainly not exactly a high-margin product...
I am conceding folks in Santa Clara apparently have some sense of humour at least. A product having no meat at all, what codename would you use? Shell? Skeleton?
Shelton! Here we are...
K.
neye eve
.net is as an integration approach. I look at the replacement of Baan by SAP as the other side of the very same medal.
Such approaches carry very high potential for improving operational efficiency and high potential of cost-reductions if you get it to work as intended, but complexity is challenging.
K.
p.s: I just upped SG&A forecasts for the next two years, projects like this one are not exactly cheap to have.
Add APM 3.0 to be intended for deployment in Fab36 (and probably in all other fabs as well), hey, the folks at least have some idea how to keep a lot of their chips busy
p.p.s: Good luck for those involved in the projects. Folks here count on your excellence.
New backend software for APM
http://www.camstar.com/
Folks are apparently working hard on improving operational efficiency. Hopefully transition will go smoother than from Baan to SAP.
K.