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.
elmer
The top of the bell curve is usually at about 70% of the highest speced number
Interesting. I did not expect that much variation of current, which is the variable here, right?
K.
elmer
I can see these numbers.
But then we have numerous reports on temperatures which do not correlate to these numbers. Ok, I dont believe any temperature readouts from monitoring tools anyway.
But then we have a lot of reports that the heatsink of Opterons only get handwarm. Ok, most of them from lower clocked CPUs.
Back to these numbers. Maximum numbers, ok. But not very far from the upper edge of Thermal envelope, it seems.
In essence, it looks like benefits from SoI are not at all low hanging fruits. Which is not exactly surprising - IBM went through all this couple of years ago already.
K.
wbmw
It's late 2005, but it's not 65nm? Perhaps this product would be K9, then?
Rather 90nm. Preferable to use a mature process for making dual-cores because low-voltage is required. AMD is not exactly good at LV - there will be a better chance to yield that from the end of the 90nm node than the beginning of 65nm node, I guess. These parts will be made for serverspace first hand, so diesize is not crucial. Bill mentioned two cores will share the cache. K.
blauboad
MS is working overtime on the WinXP port as much in order to help them succeed
Well, I guess it is in MS's very own, very best interest:
Though better their OS will support AMD-64 architecture the more lost terrain in serverspace to Linux might be conquered back. (underlying suspicion: Linux64 seems to be a quick, but not exactly well performing port).
Maybe someone with Linux-64 - insights could comment on this speculation?
K.
chipguy
The AMD boys in Austin still have to learn all
the funky switches in PGI I guess.
Guess its more due to Portland guys are apparently working day and night on the thing, releasing minor updates every couple of weeks.
K.
p.s: On a second thought, possibly Versions 5.Oa-z are just alphabetical explanations of the switches for the Austin boys.
wmbw
Thanks for correction of 3000+ pricing-relation.
By the way, how do you find this pricing extraordinarily noteworthy
I receive the message from it that AMD is capable to deliver five-digits A64-volumes immediately and six-digits-volumes in Q4. Enough to populate the mainboards out and enough to satisfy the demand from early adoptors for moderate prices.
(Plus, silently correcting XP3200+ and XP3000+ TPR )
There is more to be interpreted from the implications from the process-side of it.
Giving up XPs 3xxx allows AMD to run its bulk-process for Bartons less aggressive - which should improve yields for Bartons significantly. Which allows to phase out Thoroughbred very soon - if not immediately or even already done.
Which effectively makes a trading up path for AMD. Which is essential to make a capacity constained manufacturing profitable.
K.
wbmw
Did you notice Athlon64 3200+ is introduced with a lower price-tag than its XP3200+ - Model and Athlon64-3000+ (DTR) as well below AXP3000+ ?
K.
wbmw
"He (Ruiz) left unchanged the company's third-quarter forecasts and declined to say when AMD is targeting profitability."
Dont expect Hector or other AMD executives to make overly positive or optimistic statements before end of January 2004. Look at the terms of AMDs ESOP-exchange programs to understand why.
From what I have seen today (apart from a rather boaring launch event) the pricing for Athlon-64 products indicate a mature SoI-process already which will certainly lead to a steeper migration path as I expected it until yesterday.
As for Q3 results, I hope you will be aboard for Pattricks Contest again.
K.
wbmw
Any collapse in microprocessor demand is going to affect AMD far more than Intel, just as it has in the past.
That is certainly true for today - and will remain so until AMD will be firmly established in corporate markets and in AMD-64 brand-awareness.
The recent decade in civil aircrafts market could well be the scheme for CPU-market of the upcoming decade. But hey, we dont know if Concorde would have made as a Boeing-airplane. K.
wbmw
Any collapse in microprocessor demand is going to affect AMD far more than Intel, just as it has in the past.
That is certainly true for today - and will remain so until AMD will be firmly established in corporate markets and in AMD-64 brand-awareness.
The recent decade in civil aircrafts market could well be the scheme for CPU-market of the upcoming decade. But hey, we dont know if Concorde would have made as a Boeing-airplane. K.
doug
Are you saying they no longer sell 1.4GHz Opterons.
No. But I hope they dont get these parts out of current manufacturing anymore, but sell it from inventories of earlier manufacturing stages.
What I'm saying is that one can't really conclude anything from the offered speeds of merely the Athlon64 and Athlon64 FX-- one needs to also consider the offered speeds of the Opteron part, since all 3 are produced from the same die on the same process.
I guess "the same process" could be easily a misunderstanding. This is a dynamic thing, that is why i mentioned learning-curve. The same process is subject of tweaks every day - to be improved over time.
What came out from the same process of T-Breds is basically all the same process - tweaked by several design- and lithographic steppings as well as process parameters. Ranging from 1700+ to 3200+ CPUs within 18 months.
Same applies for K8 (and every semiconductor-process)
Does that make sense?
K.
blauboad
Its the same die. The memory specifications that can be handled by the on-die controller have been fixed some three years ago (for a planned launch date in 2002). That was before Dual-DDR came into the equation. In fact, Opteron (and FX) are not Dual-DDR capable, but 128-bit-wide-access capable (which gives pretty much same results as dual 64-bit-access) for ECC memory. 64-bit wide access (single channel) is possible without ECC.
Maybe somebody else can correct/add/clarify in a more technical way. I know in fact it is not 128-bit-access, but 1xx bit, so it is a simplified view on it.
K,
doug
They are releasing some of the parts as 1.4GHz Opterons.
As I said, hopefully not. The process should have improved within six month.
but there's nothing to suggest it's a particularly narrow range, when the output is being used in parts that range from 1.4GHz to 2.2 GHz, right?
No, not right. And yes, there is something to suggest it is indeed a narrower range than what you predict:
1. A learning curve for the industry.
2. Yield management systems to control process parameters tough enough to expect bins with low standard variations.
K.
wmbw
AMD quickly going out of business is one possibility
Well, not completely impossible. However, commodization implies price/performance and other hard factors will be dominant criteria for buying decisions. Brandings are pretty irrelevant in typical commodity markets (like Soy beans and pork bellies). Would bode extremely well for AMD.
But many many people capitalizing on status quo dont want this to happen. Lessee....
The last I would expect is a sudden death. Rather an outcome like we saw in the 80th in this market. Or what we see between Boeing and Airbus.
K.
wbmw
Why not get a presentation projector and a TV card? Why would anyone buy a big screen TV anymore?
Beamers are still expensive. Not only to buy - you need a 200 Dollar-bulb every couple of months if you use it on a daily basis.
K.
doug
Same die as in Opteron.
Sure.
So the bin isn't all that narrow.
How do you know?
It goes down to 1.4 GHz.
Hopefully not. Thats what came out half a year ago.
OTOH, it is save to say there is nothing coming out right now which would allow to be specified @2,4GHz - because FX would not have been necessary in this case.
Optimistically, 90% are at 2 GHz, 5% @ 2,2 and 5% @1,8. (That is if yield management systems work exactly as they are advertized. Which I doubt.)
Ok, realistically 90% are 1,8 and 2,0 binsplits and 2,2 and 1,6 are 5% each.
Tomorrow's imminent update of AMDs pricelist will give us a better idea of what current binsplits are.
K.
wmbw
I don't think you'll be seeing OEMs creating server product lines based on them.
I dont think either.
My point is more about perception of Intels Xeon-line. A product launched as a premium-serverproduct devalued by offering it to the consumer-market just weeks later...
And yeah, sure, AMD does exactly the same..
Well, they both are well aware of the halo-effect - which is the obvious reason boths are willing to sacrifice differentiation.
Hmm. What can we expect from accelerated commodization?
A lot of semiconductor-economies is based on the perception that IT-people do things that nobody else can do - which is difficult to maintain when world and dog can see that CPUs these people use are sold to teenagers at Best-Buy.
Lol, at the very end, this could eventually play for Itanic.
K.
InternetPlay
Oh yes. NY Times is always interesting. Btw, here is another tidbit from this article:
Advanced Micro should be able to nearly double the capacity of its microprocessor plant in Dresden, Germany, by moving to technology for making chips with smaller circuit sizes over the next year. By 2006, however, it will need a new factory.
I would guess stockbrokers will get a lot of phonecalls during the next hours....
K.
JerryR
Intel's decision to take a heavily-cached Xeon and let it go slumming with gamers is a pretty brilliant tactical move against AMD.
First, it is clearly not an active move, but a response to Athlon64-FX.
If it is exactly brillant to offer the top-of-the-line Xeon as a gaming chip, is to be seen: You know, everybody can see now it takes everything Intel has to counter Athlon-64-performance at launch. In 32-bit. I could imagine selling Xeons will be not exactly easier from this point going forward.
K.
doug
I dont think AMD is ready for HPQ K8-Volume right now. In November, they will have something for them (maybe 1 Athlon-64 for 10 Bartons?), but still not enough for a regular product line. Next year AMD will be able to deliver volumes for a regular Athlon-64-line. Maybe a good chance to take away some corporate-accounts from DELL. K.
keith, sgolds
Yeah. P4 Emergency Edition sounds about right for me for the time being. Presqwatts is what we can discuss next year.
K.
doug
...the track record of sophistry...
As I said... we all suffer from it.
Klaus
doug
Transparent FUD like that adds nothing to the board. It's just embarrassing for its poster.
Guess it is misperception, not FUD. Nothing bad with it. We all suffer from that.
Klaus
wmbw
Klaus, higher performance CPUs dissipate higher amounts of power.
As i said, I saw it coming....
http://www.investorshub.com/boards/read_msg.asp?message_id=1404618
Maybe "The downside of higher performance CPUS is they dissipate higher amount of power" is indeed the "more subtle way" the world will be "educated".
If even you find that convincing, it might work.
Klaus
AMD_Dude
but it would be really neat if they could do this for desktop & servers
Next tuesday maybe we will learn something about it. K8 has what it takes. Rumoured to be adressed as something like "Cool-n-quiet". Didn't work in earlier Opteron steppings, but is not anymore on the errata-list of Opteron.
Not sure it is already implemented in first Athlon64-systems offerings. If not, it is a matter of time.
K.
wbmw
AMDs solution for high Performance CPUs
http://www.amd.com/us-en/Processors/ProductInformation/0,,30_118_4699_7980^4703,00.html
Next generation CPUs which dont require cooling-workaround for a silent PC. Pretty boaring, aint it?
Intel has clearly the superior solution for high Power PCs. I saw it coming....
http://www.investorshub.com/boards/read_msg.asp?message_id=1404618
K.
keith
Tremendous impact. Would mean transition to 90nm at about the same time in serverspace (if AMD can deliver to their roadmap, that is).
This is something the IT-folks will certainly make thinking a lil bit. They have been told again and again Intel will always be ahead.
Where do you have it from? K.
wbmw
If Intel's 90nm chips have little advantage over their 130nm chips, then I am inclined to agree with you, but so far, I see you and others already declare leadership to Opteron without even considering what Intel has on the roadmap between now and 2005. We know very little about how Prescott and Dothan will perform. I find it hard to believe they will have nothing to offer.
Well. Starting from the last one: Sure, Dothan and Prescott will perform better than their predecessors. Not sure its little advantage or lil bit more than little. But thats beside the (my) point. My point is AMD seems to have a two years-opportunity to establish a position with its 64/32-bit USP products without being crunched by means of dual-core CPUs.
As for you and others already declare leadership to Opteron without even considering what Intel has on the roadmap between now and 2005
Leadership would require all tier-one selling it. Cannot happen anytime soon. Even if DELL would consider that, AMD would not be able to deliver volume for DELL. Not this year and not next year. 2005 - eventually. But not without additional capacities.
K.
wbmw
I don't think Pentium M is dual processor capable (pity)
Obviously not. I expected Dothan to be DP capable to allow dual-core CPUs already next as a convincing Server-CPU.
Now, Paul stated today there will be no Intel dual-core CPUs before 2005.
This imo has tremendous impact on the competitive landscape in 2004/05: The window seems to be open long enough for K8 to establish positions in serverspace, Workstations and High-Performance Desktop.
12 Dollar per share now looks dirt-cheap for AMD common.
K.
elmer-
Really good to have people here teaching us how manufacturing chips really works.
Thanks.
K.
Golfburn
I am talking about three HTT-channels to be tested to determine the good ones to be connected to pins in the final assembly - depending on which Opteron/Athlon-model the die is good for.
What is so funny with it?
K.
elmer
You really think it is necessary to know how exactly they do it..
That is interesting as well..
chipguy
I do not see any unsurmountable obstacle to prevent AMD from testing HT-channels before bonding out.
I have no idea if that can be done at wafer sort or requires an additional step (maybe even an additional tool) in a later stage of the backend process before bonding out.
K.
chipguy
Being unable to test before bonding out would make the whole concept of redundancy in chip-design obsolete.
I am pretty sure the K8-design allows using any of the three HTT-Channels for an Athlon64 or Opteron1xx and any two to make an Opteron 2xx. (Not only) AMD depends on the economies from such designs.
K.
sgolds
Nice to see, but dangerous to use as a data point for investment
Well. Guess statements like this are exactly what could ignite and fuel hype. As humble as that.
K.
chipguy
Thanks lot for the explanations. As far as I understand there is some redundancy in the hypertransport channels: Only 8xx Opteron-models need every HTT-Channel to meet all specs.
If one channel does not meet all spec, its gonna be a 2xx-Opteron, if two do not meet specs its an 1xx or an Athlon64.
Parametric yield loss would only occur if all channels fail to meet specs - which hopefully is not a significant number.
K.
chipguy
- high speed I/F needs new ATE investment; likely source
of non-trivial parametric yield loss
Could you pls enlight me what that means?
Thanks. K.
"However, according to a North American reseller of server systems, the Opteron is selling better than sales of Xeon and Athlon MP chips put together."
http://www.theinquirer.net/?article=11531
K.
keith
maybe more desinformation.
According to an email buggi received some time ago, this OPN indicates Vcc of 1,5V. K.