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.
Yourbankruptcy -
EP, I beleive Madison has faster clock frequency and bigger cache. We can roughly estimate how well it will go. There is no difference is scaling, right?
Why assume that? Larger caches mean better scaling.
A more important point is that we have no facts at our disposal here, only speculation. I do not accept any claim whatsoever about Hammer performance before the fact. The marketing claims of a company on the verge of bankruptcy is not a reliable source. When there are public benchmarks that can be verified, and volume production then we can talk.
YourBankruptcy -
As Opteron is borrowing a lot from EV7 and is made by the same people, you can expect 32-way Opteron to deliver very similar numbers.
Next year's Opteron system verses what will very soon be yesterday's Itanium. You're going so far off into fantasy land that there's little point in continuing this topic.
EP
Yourbankruptcy -
The difference between you and me is that I already know than 4P Opteron 1.6 Ghz is faster than 4P Itanium II 1 Ghz, even though single Opteron is slower than single Itanium II, but you will learn that four weeks from now.
You're certainly painting with a very broad brush. I'd like to see us refrain from making claims we can't back up, and you can't back that one up. Who knows what wbmw may or may not know that he's not privileged to discuss?
EP
Paul -
but anyone who tries to represent Athlon as anything but a resounding success for AMD(albeit with a few snags along the way), is just being dishonest.
It depends on how you define success. AMD has lost money, overall, on Athlon. They've had to sell off profitable divisions to continue to fund it. Their share price has suffered and their debt is crushing. Staggering losses are followed by more staggering losses. Their cash position is degrading and there is no relief in sight. This is not a "resounding success" in my book. In my view, this is the measure we should be using when defining success.
I'm confident Opteron and A64 will do the same
You might be right.
chipguy -
Unfortunately your claim about Opteron looking really, really good seems at odds with the fact that AMD hasn't snagged any first or second tier OEM as customers for their chip. What's up with that?
The question seems to be, if all the tier1 oems have looked at Opteron and nobody has said "yes", is that the same thing as them saying "no"?
wbmw -
Any AMD investors on this board?
Yes, I'm here. I happen to believe money can be made on AMD by understanding that AMD will always return to these levels. They have for 20 years. I also agree with your prediction that AMD will miss their estimates for Q1 and post a larger than expected loss. Now, how can you make money if this prediction plays out? I am long AMD but short April $5 and $7.5 CCs, plus I'm short April $5 puts.
IBM breaks silence on server speed
By Stephen Shankland
Staff Writer, CNET News.com
March 27, 2003, 11:53 AM PT
update IBM posted long-awaited performance scores Wednesday for its x440 server with 16 Intel Xeon processors, a system Big Blue is using to try to impose itself on a key part of the server market.
The system was able to perform 151,000 database transactions per minute, according to the scores. That's ahead of a rival 8-Xeon Hewlett-Packard DL760 G2 that clocked in at 115,000, but still behind the 32-Xeon Unisys Orion 230 that racked up 234,000 transactions.
The $1.7 million IBM system's score was about one-third of that posted by the highest-ranked system overall, a huge 128-processor Unix server from Fujitsu that comes with a $12 million price tag.
The benchmark, or speed measurement, test was designed by the Transaction Processing Performance Council. Server makers can spend millions of dollars preparing for the widely watched TPC-C test. Although the benchmark results can be somewhat inflated through careful tuning and the use of unusual storage systems on the part of manufacturers, a good score can't be achieved without a good system.
IBM's x440 is the spearhead of a campaign by the company to conquer the market for Intel-based servers. For years, the capabilities of Intel servers weren't far removed from those of ordinary PCs, but IBM, HP, Dell Computer, Microsoft, Intel and component maker ServerWorks have been coaxing the systems' features closer to those of more powerful Unix machines.
Market research firm Gartner expects that Intel-based systems, with an projected $20 billion in sales, will make up the largest part of the server market in 2003. Compaq Computer's ProLiant line, now sold by HP, has long led the market. Dell is gaining increasing share through strong price competition, while IBM is angling to get ahead by applying its considerable server engineering skills to otherwise comparatively ordinary machines.
The IBM server that racked up the TPC-C result used Microsoft's coming Windows Server 2003 operating system, scheduled to arrive April 24.
That operating system is the first from Microsoft to include non-uniform memory access, or NUMA, designs. These designs, which are widespread in higher-end servers with dozens of processors, vary the time needed to retrieve data from memory depending on how close that memory is to the processor that needs it.
IBM has its sights set in particular on Dell, hoping to compete better on price through a deal under which manufacturer Solectron now builds many IBM Intel servers.
Big Blue is touting two recent gains against Dell. Last week, IBM announced that the Asian Art Museum in San Francisco swapped older Dell systems for an IBM iSeries machine that includes eight IBM Intel servers handling tasks such as online ticketing and museum membership.
And Friday, IBM plans to announce that it displaced Dell at Online Taxes, which is using seven lower-end Intel servers for 150,000 customers filing taxes online. IBM's Global Services division is hosting the servers, Big Blue said.
UpNDown -
Thats why you can buy an Opteron server and run your existing Windows 2000 or 32-bit Linux server applications on it.
Really??? Where can you buy an Opteron server other than in your dreams?
Bull has posted SPEC scores for Itanium systems.
http://www.spec.org/osg/cpu2000/results/res2003q1/
Semi -
Sound like anyone we know????
Sounds like almost everybody we know.
wbmw -
If not, then maybe the author at the Inquirer is just blowing FUD.
Please!!! I'm not that kind of guy....
JoeP
I think he was being sarcastic
Spokeshave -
The recall remark was intended to be tongue-in-cheek. I think most people got that.
I realize that. I just remember being chastized for doing essentially the same thing. I wonder if that person will do the same with you?
I don't think so either.
Spokeshave -
Another possible Intel chipset recall?
Pretty low Spokeshave. No data whatsoever, just an accusation.
YB -
I was gambling that monday price will stay at Friday level and lost :(
So why not put in a good until cancelled limit order?
Yourbankruptcy -
I agree, as Intel can't beat AMD on technical benchmarks, it clearly makes sense to compare financial benchmarks instead, if you need to make Intel to look better.
Nice to see that you agree with yourself. Not a hard one to predict. As I said before, it is tempting to show the overwhelming number of benchmarks that show otherwise but it is getting a bit boring. I posted something about buying AMD shares and writing CCs. Why not change to a subject we can all agree on. Making money.
Bought some more AMD yesterday.
Wrote April $7.5 CCs.
EP
YourBankruptcy -
Athlon beats P4 in 3DMark score:
As tempting as it is to point you to benchmarks that show a completely different result, I'll forego that and point out that the only benchmark that really matters is the one on the bottom of the companies quarterly financial statement. Intel's is black and AMD's has been and will continue to be for some time, red.
wbmw -
I don't know if I should bother to dignify this blatant lie.
wanna, this is why I put the guy on ignore years ago. He's a liar and always has been. Why waste your time with him?
Only a fool argues with a fool.
Spokeshave -
You are assuming that all production from that 1 to 2% of capacity started between now and September will somehow all result in completed units in September.
This topic has been discussed several times and I've said that this assumes wafers are coming out now. If they don't have any wafers now, they don't even know if the device works on this stepping.
Spokeshave -
Philo obviously falls into the agape catagory. (er, that *is* what you meant, right Elmer?)
Rest assured :)
sgolds -
eros - erotic love
philo - friendship (male)
agapee - devine love
all phonetic. I don't know how they spell out in english.
UpNDown -
I'd be interested in hearing your estimates on what percentage of AMD production must be devoted to Athlon-64 right now for them to make a successful September launch.
What assumptions should we make about yields and performance/binsplits? I would assume the yields must be at least acceptable or why bother, but the same holds true for performance. I showed before how AMD could have 100K in oems hands by the Sept launch using only 1-2% of their capacity if Athlon64 is in production now. Assuming they are coming out of the fab as we speak, 100K units in 6 months is less than 4K per week. For a 104mm2 die, at only 100 good die per wafer that's just 40 wafers per week! That's really bad yield and still they easily meet the volume release with fewer wafers per week than they probably break. Capacity can not be a problem if both SOI and Athlon64 are even marginally healthy.
No need to take my word for it. You can verify this for yourself. You know AMD's fab capacity. You know the die size. You can calculate yields yourself and punch them into your calculator. Some people here are bending over backwards to find some way to explain away how a 5500 wafer per week fab is somehow really a 2500 wafer per week fab but it doesn't wash.
EP
Spokeshave -
O philo moo
My friend.
Sheesh.. And you're the one who was translating Greek...
Spokeshave -
The word "penultimate" loosely translated means second place
Don't quit your day job O philo moo.
Sgolds -
I agree completely. If we can pull this off with minimal loss of life we will score an enormous humanitarian victory. This is the best possible outcome for a bad situation.
EP
Spokeshave -
I am not an industry insider, and I do not know the arcane lingo. I coined the phrase in an attempt to describe the types of defects that I believe that you are not accounting for that affect the yields initially when a new core is introduced. Forgive me for not using the correct terminology, but please try to consider this from a more conceptual standpoint.
I grant you this. I'm not even sure if we're arguing here because my estimate of AMD using 50% of their capacity was in reference to their existing products on their bulk .13u process. Opteron/Athlon64 will not fit into this category but Barton will. Are we in disagreement as to which category Barton falls into? If so then please provide some explanation as to why Barton would have some unique process aspect or sensitivity that doesn't affect their other higher volume bulk products.
Spokeshave -
The ones I referenced, and others, are directly related to problems with masks, number of layers and other things that accompany a new die design.
But there are other designs and steppings running on the same process. They have masks too and use the same number of layers and all must follow the same design rules defined for that process. The size of metal lines must meet the requirements for the current they carry. The spacing between lines, and a whole host of rules that I'm not up on are process specific, not design specific. Sure sometimes the design can be faulty, even if it's bug free, because of process issues that need to be accounted for, but that design would never go into production so only a small number of wafers would be wasted. This could very easily apply to Athlon64 but it would apply to Opteron as well. One of the rules for releasing a process into production is usually demonstrating it can run in volume at a defined defect density level. Maybe SOI hasn't demonstrated that ability yet?
There are three relatively new products that are either in production or tooling for production - Barton, Opteron and A64. Each is a new die design and each will have a greater design-related defect density than they will at the end of their production life.
The defect types you described are obscure and not a significant contributor to yield loss once a process is in volume production. They are the kinds of problems that are worked out before a process is considered "manufacturable". In 20 years I have never heard anyone mention "design related defect density". I don't even know what you are trying to say. Intel runs several products on their .13u process. P3, P4, Banias, and soon Itanium (if not already) and their yields match predicted levels based solely on die size and the defect density for the process. There is no such thing as a design specific defect density. I have worked with other foundries and the same rule applies. Tell me the die size, the current defect density for the process and I'll tell you the yield you should expect. I never need to know the design. The yield calculator I gave a link to didn't ask you what design it was did it?
Spokeshave -
You have won the prize for most wrong points in a single post. In fact I can't find anything in there that's correct, including your claim that I claim to be a process expert. I have stated many times that I am not a process expert. My expertise is in manufacturing (post fabrication), test and design for test. The overwhelming number of defects can be modeled and predicted and are simply not design specific despite your claims that they are. Design rules are in place to ensure that designs are not subject to their own unique failure modes to any significant extent. For a mature process, most defects are particulate in nature with a smattering of other more obscure failure modes. Early on you will see more exotic failure types but those would be worked out by now for AMD's .13u process. That's why it is easy to calculate how AMD can produce the current demand for their bulk .13u products with half a fab, unless there is something very very wrong. I'm giving AMD the benefit of the doubt and assuming their .13u process is running well.
Sgolds -
Fab 30 has generally been around 50% utilization for shipping product for these reasons. The other 50% is busy preparing the future.
You make an excellent apologist. I don't buy it.
Spokeshave -
As with most production processes, yields start off lower and increase as the process matures, yet you are assigning a mature yield figure to all processes. I am certain that the overall weighted average yield for the fab is quite a bit lower than the figure you use. It is completely unreasonable to assume that all concurrent processes are producing excellent yields.
I'm not assuming any such thing. Allow me to say it a different way. AMD is capable of supplying the demand for their .13u Athlon derivatives with half their fab capacity, assuming good yields and binsplits. That is a mature process and runs mature products. While Barton may be a new version of the design, the defect density doesn't know what product is being produced on a wafer so for a mature process the yield is more a function of the process, not the design.
This leaves about half the fab's capacity available. As I showed before, AMD could produce about 100K units for immediate delivery at September "release" of Athlon64 by using just 55 wafers a week with normal yield or 100 or so if poor to bad yield. Yet people here claim immediate availability won't happen. Why? It would still leave about 2700 wafers per week in capacity available for process development and other forms of unproductive activity. AMD still has a development fab in Santa Clara/Sunnyvalle so process development goes on there as well.
The point is that if Athlon64 is not in production today, it can not be because of lack of capacity and whatever problems prevent if from being manufacturable at this time would likely affect Opteron as well. Just a guess on my part, I admit, but that's my take on it.
EP
Paul -
could you share what facts you are basing the 50% utilization figure on?
It is based on 5500 wafers per week capacity at F30 and various die size.
You can calculate their capacity for yourself but using this yield calculator
http://www.icknowledge.com/misc_technology/die_calculator.xls
Murphy model is fine to use. 0.25/cm2 is considered excellent yield.
You can find die sizes at various sites, or from google.
See for yourself what AMD can produce, assuming good yields and binsplits.
Yourbankruptcy -
EP, why do you think that the problem is with the chip itself? Do you accept the idea that the problem may be with AGP bridge or just plain software, Microsoft problem?
I don't really know. I would only be guessing anyway. All I know is that if they are 6 months away from turning on the tap for Athlon64 they really don't have a clue today if they can meet that schedule and Opteron is more complex and therefore more difficult to produce so how can it be any better off?
Tonight we hear Saddam talk about crushing the American forces. I don't mean for a second to compare AMD to Saddam, but only to point out that desperate times call for desperate rhetoric. Do you consider Saddam's situation when listening to his rhetoric? I do. AMD is not evil and their management are not despots but shouldn't you consider their desperate situation too when evaluating their rhetoric? Talk is cheap and AMD has been spending it freely. Let's see what they can deliver.
Yourbankruptcy -
About this 50% initialization. Do you have any clue what they are doing there?
I'm assuming this is all the market will absord at this time.
Maybe they started making Athlon 64 is vulume, then found a bug and dumped everything?
Sure that's possible but would that bug also effect Opteron? Also remember this is an Athlon with 2 extra pipeline stages, a memory controller and a aHT port on SOI. They've had over a year to find and fix bugs. It's not like an Itanium which was a whole new design from the ground up. I can't believe the design was that hard. Process continues to be the likely problem as everybody in the world predicted from the start. Sure, they've called in the cavalry (IBM) but where is the evidence to show that IBM has ever established good yields on their own process? They don't have to ship high volumes. Who knows what their yields are?
Maybe they started big flash 0.13 ramp?
That's Fab25 in Austin.
This secret production has to be related with something unannounced.
What secret production? XP and Barton demand only requires half a fab, assuming good yields and good binsplits.
Do I undertand correctly that cost of running the fab is not that much different with lower utilization?
I guess a large portion of the costs are fixed but profits won't come from half a fab.
God Bless America.
Sgolds -
You can embrace whatever theory you wish but so far it's just a theory. You have only your hopes to sustain you. I have my own ideas too. I see a modern fab at only 50% utilization. I see a product release scheduled for next month but a simpler version of that same device that won't even start shipping, if you believe some people's claim, for 6 months. With half a fab unutilized it can't be capacity that's delaying the simpler device, Athlon64. Something else is wrong and it's reasonable to question how the simpler device is 6 months away from production but the more complex device is ready to go. I don't believe it and I've been watching silicon go into production for over 20 years. You can believe AMD's promises but I'll trust my instincts.
EP
yourbankruptcy -
Why do you claim that this year both Opteron and Athlon 64 launches will be paper?
I don't know that either one will be vapor but I'm guessing that Opteron will be. You should be arguing with the others here who claim that the Athlon64 launch in September will be only the start of production flow. That's vapor. Why do you give those guys a pass?
Edgarcayce -
Mr Phud-Are you enjoying the AMD rally to date? Looks
like it will be interesting from here on with the Opteron and then the Ath 64 coming later this year. R U still in?
Yes I'm still in and plan to stay in for as long as I can play my options and still make money. I'm happy for you guys and I hope you make a ton of money.
EP
sgolds -
Elmer, evidence: The official release date of Opteron is April 22. Now, you know the lead times in production. Thus, in order for there to be demonstratable product in systems (by Newisys, for instance) on April 22 then there has to be manufactured product well before now
Ok, so you have no evidence of production material in the hands of oems, only AMD's promise of a release data which may be vapor. Others here have pointed to Athlon64s September release data and insist that "release" means that's the day AMD starts shipping production material but you say "release" for Opteron means they have been shipping production well before release date. I wish you guys could get it straight.
but to expect an announcement with this buildup but no product - that sort of expectation would be incredulous.
A release without material would be normal for AMD, not an exception. They are the Kings of Vapor and I don't expect Opteron to be any different. I'm truly amazed that you believe AMD's promises at this point.
EP
sgolds -
Yes, the OEMs do know the difference: They have production Opteron processors in production boards, and I'm sure they have very reliable benchmark data.
Please provide data to support this claim.
leaked benchmarks show that it scales very nicely and is more than competitive with Xeon. Soon enough those benchmarks will be official! (Actually, Opteron looks competitive with Itanium 2 also, but that's a whole different market...)
The same old claim. Rumors of future Opteron verses today's Intel products. How will real silicon compare to tomorrow's Intel products? You don't know.
Article on Springdale with benchmarks
http://www.xbitlabs.com/articles/chipsets/display/springdale.html
The launching of Canterwood, Springdale and Pentium 4 processors with 800MHz bus should become a significant move forward. As we see, speeding up the Pentium 4 bus together with the shift to faster memory improves the performance very noticeably. NetBurst architecture implemented in Pentium 4 turned out perfectly scalable with the FSB frequency growth. Therefore, as soon as Pentium 4 processors with faster 800MHz bus come out, the performance of top Pentium 4 models will get to a totally new level.