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.
Dew -
Are you disagreeing with Alan that this rule of thumb has any value for semiconductor-industry investors and analysts?
No I'm not disagreeing with him, just pointing out that what a process is capable of producing and what the market is willing to buy may not line up.
Assuming other variables are kept constant, by what magnitude would you expect the frequency distribution to widen at a foundry relative to a dedicated house? 50%? In other words, if Alan’s “+/-10%” is a reasonable rule of thumb for dedicated houses, would “+/-15%” be a comparable rule of thumb for foundries? T.i.a. Dew
I've never analysed the frequency distribution of a foundry because all the foundry based products I've dealt with had only one speed, but I would "expect" the numbers you quote to be reasonable. My point is that in a market where frequency is so dominant it is a mistake to assume that the market sweetspot corresponds to the process sweetspot for the entire industry. Is the entire industry making money or only one player? Is it possible that one player may have a process advantage? Once again getting back to AMD because it makes such a good example, they are having real problems delivering meaningful quantities at higher speeds where APSs are better. Their output is very poor for a fab their size, their ASPs stink and they're losing a ton of money yet their marketing says they have 2800+ product. Do they really? When the market leader has a distance on the rest of the field you will find everybody elses marketing teams making claims they can't deliver to and you should take that into consideration when judging where their fab's speed distribution lies. TMTA may be selling 1GHz product because the market demands it but that doesn't mean the binsplits are good enough to allow any profits at that speed. How many wafers do they have to go through to get enough of those 1GHz parts? Is market share more important now than profit?
EP
Dew -
Before I decided to answer this post I looked at the other responses and I think they both missed the mark. First off, you can not infer anything whatsoever about speed distribution based on maximum frequency rating because market forces dictate which speeds can be sold and at what price. TMTA may simply be doing an "AMD". Announcing products it has no ability to manufacture but must give the impression it can. If however you want to assume that 1GHz is at the high end of the frequency distribution then you need to know something about who is fabing their parts before estimating the range of distribution. A foundry is optimized for yield and not speed. I would expect to see a greater frequency distribution from a foundry than a dedicated CPU house like Intel or AMD. Depending on the forces brought to bear, the later two would probably have a much tighter frequency distribution for a given process. Perhaps one of those two may even suffer major yield loss trying to raise the distribution to remain competitive. I will not speculate on which one that might be but I will point the reader to the recently released financial statements where perhaps clues may be found...
I would also discourage anyone from concluding that the pricing and volumes for any speed grade is a surrogate for binsplit determination. This need not be the case. Once again market forces could determine the best pricing points for maximizing profits when binsplits may be well in advance of the competition. A device labeled xGHz could perform well in excess of that speed. The rating simply guarantees a minimum. If the market sweetspot happens to favor a point below the distribution mean than a faster device could be substituted. Too many unknowns and too many variables to draw conclusions on TMTA but for the other two companies Intel seems to have no trouble supplying their highest bins while AMD appears unable to supply theirs.
EP
119 -
You came on here and made insulting comments and now you're crying poor me. Get over it. You've got a chip on your shoulder and got the same thing you dished out. Get over it and make a meaningful post.
EP
119
I'll let you investment experts get back to business while I go pick up some more chicks.
Don't let the feathers mess up your car.
Smooth -
What are they doing, eating them?
Could be. They didn't sell them, at least not all of them. 2 for 1 deal? It was something other than sales.
EP
Wbmw -
I thought you were asking me if you should sell CCs. That's the question I was answering. Sorry if I misunderstood. I won't try to sway your decision but if you do decide to sell CCs I think you can expect to generate a cash flow considerably in excess of what the cash would do just sitting around. I expect AMD to revisit $7.50. As you know AMD is a religion or maybe better called a cult. It has it's devotes and they will stand in line to throw their money away just to show their devotion. This will give good options premiums. Meanwhile the case can still be made that Hammer may be a viable highend product. The ASPs that can be generated by even moderate volumes can at the very least give AMD a infusion that will keep them alive long enough for those options premiums to lower your cost basis considerably. My goal is cashflow and I think this is a very good candidate.
EP
Anyone know what this is about?
UMC wins Intel unit Primarion order for power supply chips - report
TAIPEI, Jan 16, 2003 (AFX-Europe via COMTEX) -- United Microelectronics Corp (2303.TW) has secured an order for power supply chips from Intel Corp unit Primarion, with shipments due to begin in the second quarter, the Economic Daily News reported, without citing sources.
The order is expected to enable UMC to use up its capacity for 20,000-40,000 8-inch wafers per month, it said.
wbmw -
Should I buy back at a lower price?
No. Just let them expire and do it again.
EP
Wbmw -
I have said all along that I don't expect AMD to go anywhere so that's why I sell the CCs. I just sold some Feb $7.5s today for only $0.50 but I didn't want to wait until after earnings (Ha!) because anything could happen. This is much much worse than I expected. The higher speed TBreds didn't help ASPs very much did they? I expected the operating loss to be much less...
Still I think AMD will hover in the $5-$7 range. I'd just sell the $7.5s if I were you. What's going to happen in the next couple of months to propel this stock higher? I'll be very careful with the Aprils though because that's just after the next earnings (Ha!) report and AMD obviously refuses to issue warnings.
Do you think it was just a coincidence for that Hammer benchmark to appear today?
EP
Wbmw -
Elmer, it looks like AMD is also claiming "record unit consumption of PC processors", whatever that means. Yet, if you look at their sales numbers for the CPU division, they are still down 40% from last year. How do you have record unit sales, higher ASPs, and less revenue? This stinks of bullsh!t
Looks like they carefully chose their words. "Consumption" is not the same as sales.
EP
http://biz.yahoo.com/bw/030116/160411_1.html
AMD Reports Fourth Quarter and Annual Results
Thursday January 16, 4:30 pm ET
AMD Increases Sales by 35 Percent and Narrows Operating Loss From Ongoing Operations by Approximately $100 Million
Sees Record Unit Consumption of AMD PC Processors
SUNNYVALE, Calif.--(BUSINESS WIRE)--Jan. 16, 2003-- AMD (NYSE:AMD - News) today reported sales of $686,430,000 and a net loss of $854,740,000 for the quarter ended December 29, 2002. The net loss amounted to $2.49 per share. Excluding the effects of restructuring and other charges(1) to show the company's results from ongoing operations, the fourth quarter net loss was $235,145,000, or $0.68 per share.
<more>
Fingolfen -
Of course, far too often journalistic integrity goes out the window to "sell product".
Yah. They "market" their product [journalism].
EP
Fingolfen -
I've noticed a lot of really shoddy journalism surrounding benchmarks and the K8 lately.
Consider their situation. Who would be interested if they said wait indefinately for mediocre performance?
EP
Fingolfen -
Any word on the level of optimization for I2 / Opteron on this bench? What OS? Beta patch? System configuration?
No more info than what you've seen.
One other thing though is the Itanium2 score was an old one. The latest I2 score is 600 also.
http://www.sap.com/benchmark/pdf/cert6302.pdf
Fingolfen -
Given that the fastest hammer out in serious public view at this point is a 1.4GHz part with underwhelming benchmarks, I honestly don't know that Intel's the company that needs to do the catching up at this point
There is a new benchmark out for Hammer that shows an impressive SAP score. I'll attach an English translation.
http://www.heise.de/newsticker/data/as-16.01.03-002/
Already early prototypes of the AMD of Opteron processor knew in four-fold systems the met competition -- all the same whether 32 or 64-bittig -- with the SAP SD bench mark in the shade place, and with only 1.6 GHz clock (2 GHZ are planned) with the Launch. With the SAP R/3 Sales and distribution bench mark (SD) becomes typical Business transactions (orders to give up, etc. corrects.) accomplished in a large data base. As measure serves the number of the SD users, who can be served within a given time interval.
Bench mark for different hardware configurations, approximately for "2 animal 4way servers" are led individually. For this for example HP for the Itanium indicates 2 in the rx5670 470 SD user . First, still few optimized attempts a few weeks ago were with the quadruple Opteron system with R/3 and DB2-Datenbank under 64-Bit-Linux already with 550 SD Usern, which were served of middle period of reply with 1.9 seconds, about 600 SD Usern lay the time still scarcely over 2 seconds. In the meantime the 600-SD-User-Marke was probably exceeded. No bad prospects thus for AMDs entrance with servers with Opteron processors. ( as /c't)
Intel accelerates Itanium schedule
By Michael Kanellos
Staff Writer, CNET News.com
January 15, 2003, 9:00 PM PT
Intel has changed the release schedule for its Itanium chips for servers, adding a new chip for 2004 and moving the launch date of an Itanium with two separate processors to 2005 from 2007.
The changes reflect Intel's confidence in its ability to release high-end server chips faster than competitors and thereby gain the performance high ground, said Jonathan Eunice, principal analyst at Illuminata.
Itanium 2 currently ranks with the best server chips in the market, but the new release schedule will enhance the chip's attractiveness and put pressure on competitors to step up their own schedules, something historically they have been loath to do.
Intel's "design teams and design resources are well stocked, so they can do a shrink early or do a dual-core (chip) early. They have a lot of leeway that would stress out a Sparc development team," Eunice said, referring to shrinking the size of components on a chip and to Sun Microsystems' UltraSparc processor.
Under the new schedule, the Santa Clara, Calif.-based chipmaker will release Madison, a souped-up version of the current Itanium 2 with 6MB of level three cache, this summer, according to Jason Waxman, marketing manager for enterprise processors at Intel. Increasing the cache, a reservoir of memory located on the processor, generally enhances performance.
The chip, which will contain around 500 million transistors, will run at 1.5GHz. Madison, like the entire Itanium family, is a 64-bit chip, meaning that it can digest data in 64-bit chunks (unlike 32-bit Pentium chips). 64-bit chips typically fit into the most expensive and powerful servers.
Soon after, Intel will release Deerfield, an energy-efficient Itanium 2 for rack and blade servers.
Then, in 2004, the company will come out with a new version of Madison that will contain 9MB of level three cache. Now most server chips come with 1MB of cache. This chip was previously not on Intel's roadmaps, Waxman said.
Two cores better than one
In 2005, Intel will follow with Montecito, which will contain two Itanium processor cores on the same piece of silicon. Dual-processor chips are pretty much what they sound like: single chips that contain two separate "brains" so they can best a single-core chip but cost less than two separate ones.
It's one of the hot design ideas in the chip world. IBM came out with the first dual-processor chip for the server market, Power4, in 2004. Sun's UltraSparc IV, debuting toward the end of the year, is expected to contain two processor cores.
Analysts have also noted that AMD's Opteron chip, coming out in the first half, could be redesigned to accommodate a second processor core.
Originally, Montecito, due in 2004, wasn't a dual-core chip, but it morphed into one after engineering and manufacturing teams concurred that a dual-processor chip could be mass manufactured at Intel by 2005.
"Our dual-core (chip) was originally planned for the following generation of chips," said Waxman.
Besides enhancing performance, Intel may use its dual-core chips to undercut IBM, said Kevin Krewell, senior editor at The Microprocessor Report. In larger servers, Oracle and other software vendors charge a licensing fee for every processor in a given server.
To date, IBM has said that the Power4, although a single chip, is two processors. Hence, software customers have to buy two licenses for each Power4 chip. Intel is already indicating that it will consider Montecito a single processor, requiring only one software license, Krewell said.
"I think they are going to sell it that way to make it cost effective," to switch to Itanium from other servers with different chips, Krewell said.
Montecito, Krewell added, will be made on the 90-nanometer process, which means the average feature inside the chip measures 90 nanometers. Madison, Madison II and Deerfield will contain features measuring 130 nanometers. The current Itanium 2 comes with 180-nanometer features. Reducing feature size allows companies to squeeze more transistors on a chip.
Not as sweet by any other name
Continuity is another theme. All the forthcoming chips--Madison, Deerfield, Madison II and Montecito--will be sold under the Itanium 2 name.
All these chips will also fit into the same motherboard sockets and be capable of using the same chipsets currently used in Itanium 2 servers, said Waxman. In turn, this will reduce the need for server makers to redesign their servers with each new chip release, smoothing the commercial adoption.
Despite its long and often controversial history, the Itanium family appears to be gaining momentum in the market, according to Eunice. Back in the mid-'90s, analysts speculated that the chip, designed by Intel and Hewlett-Packard, would become one of the most popular for high-end servers. However, the first version, formerly code-named Merced, was delayed several times and offered only middling performance. It finally debuted in 2001; sales were dismal.
Itanium 2, formerly McKinley, came out in July 2002 and was substantially different than the first Itanium. Analysts have given it fairly strong reviews.
Itanium 2 "competes or outperforms the fastest Alpha and Power chips. It is right up there at the elite country club of performance," Eunice said.
Still, despite strong benchmark scores, sales started slow. The economy was in a slump, which discouraged interest from customers, software developers and hardware makers.
Interest, though, appears to be growing, Eunice said. Laboratories and other scientific customers are increasingly deploying Itanium 2 servers. More software tools are also coming onto the market. If the tide changes, Intel could benefit.
"2002 was a terrible time to launch a new technology," said Eunice.
Intel may release new Xeon processor in 2004
Ashlee Vance, IDG News Service\San Francisco Bureau
January 15, 2003, 13:50
Intel Corp. appears to be readying a new Xeon processor code-named Potomac that should arrive in late 2004.
The Potomac processor was shown as the follow-on to Intel's current Xeon processor code-named Gallatin in a Hewlett-Packard Co. presentation posted on an Australian HP user Web site called OZTUG.org. While the presentation showed Potomac arriving in late 2003, an analyst and researcher at IBM Corp. said that late 2004 would be a more likely release date for the chip. Potomac will be the next chip in the Xeon MP family, which is used in servers with four or more processors.
An Intel spokesman declined to comment on unannounced products.
The Xeon processors usually have similar speeds to Intel's Pentium 4 desktop chips but include some higher end features such as larger caches that make them more suitable for use in servers. Intel also offers the 64-bit Itanium processor to server makers, although the 32-bit Xeon processor is more widely used at this time.
Analysts were reluctant to speculate on what features Potomac may have but said Intel could prepare a surprise or two by 2004.
Over the course of this year, Intel is expected to increase the cache sizes on its Xeon chip and push up the chip's speed, said Dean McCarron, principal analyst at Mercury Research Inc. in Cave Creek, Arizona. In 2004, however, Intel will be able to take advantage of a new .09-micron manufacturing process, which helps shrink the size of the chip's circuitry and which could open the door for dual-core Xeon chips.
A dual-core chip has two physical processor cores on the same piece of silicon, which makes one chip work almost as effectively as two.
"Intel is looking at multicore technology," said Kevin Krewell, senior editor of the Microprocessor Report in San Jose. "There is more mainstream software out there now that can make use of SMP (symmetric multiprocessing) in dual-core chips."
Intel could also decide to make Potomac its first Xeon chip to support both 32-bit and 64-bit instructions for its x86 architecture, Krewell said. Such a move would mimic Advanced Micro Devices Inc. (AMD), which will release the Opteron processor this year to give customers both a 32-bit and 64-bit option for lower-end servers
"I do think that Intel, at some point in time, needs to expand Xeon beyond the 32-bit architecture," Krewell said.
Intel has been rumored to be working on a similar chip to Opteron, but Krewell, like a number of industry sources, speculated that Intel dropped this project due to Microsoft Corp.'s support for AMD's 64-bit processor design. Microsoft was reluctant to support two of the so-called x84-64 architectures and sided with AMD instead of Intel, the sources said.
"AMD has staked the first claim and is much more committed to that type of architecture than Intel," Krewell said. "Eventually, I think Intel will have to say there is a future of the Xeon family that will have 64-bit extensions."
You Suck -
Could you gather your thoughts (ha!) and tell us what your point actually is? If your point is that Intel is over valued based on fundamentals then you're right but in my view that valuation basis is not the only one to apply here.
Also the articles you posted were non sequitur. The first one describes Intel's manufacturing capability and the second one starts off referring to outsourcing.
EP
You Suck -
What was INTC's previous guidance on CAPX for 2003?
I don't know the exact figure but obviously it was higher. What can we conclude from that? I think the conclusion is that under current conditions Intel does not see a need to expand capacity at an accelerating pace. The bringing online of F11x at ~200,000 sqft of cleanroom space makes it about twice the size of the competition's single fab. With 300mm wafers producing ~2.2x the number of die as a 200mm wafer it looks to me as though Intel brought online 4.4x the capacity with a single fab. With Prescott's die size being well under 100mm2 it just about doubles capacity again compared to the current generation of Intel processors, and there are more 300mm fabs. Would you be building more fabs under the circumstances? I think that somewhere you have a point you are trying to make but I can't figure out what it is. Please explain your point.
If this is just a hit & run bashing then don't waste our time.
EP
You Suck -
Either they've done it again and are ahead of plan or they don't have the visibility and are like a deer in headlights at this point. I'm voting for the truck with its high beams on <GG>
And your point is??? Just who has visibility at this point? Are we going to war in Iraq or aren't we? Are we going to fight N. Korea? Are terrorists going to do something big soon? Who the heck knows? You are a master of the obvious. Intel took a risk and expanded capacity. The economy didn't spring back so they don't need to do any further expansion at this time. You need a Ph.D in economics to figure that one out? What's so hard to comprehend here? You think you have some brilliant insight? A blind man could see this.
EP
You Suck -
I think you're missing the whole point. Intel's investments in 300mm Fabs over the last couple of years were intended to add capacity. They've accomplished their intended task. The ~220% increase offered by 300mm wafers means Intel doesn't have to spend the money on cap exp like they have in the past. The reduced cap exp is exactly why Intel invested as they did. This is going over your head.
EP
Dew -
INTC’s revenue guidance for 1Q03 is 3-10% down relative to 4Q02.
Q1 is always slower than Q4.
EP
John -
As a (SF)Bay Area native, I'm sorry to see the Niners lose but they were beaten by the better team. Hat's off to the Bucs! Let's hope the Raiders can meet the Bucs in the Super Bowl Battle of the Bays...
EP
bababouie -
Intel will make $$$$ on each set sold for many years to come...
Wonderful news. Thanks...
EP
Alan -
I just checked the entire list, and Intel does have a PIII 350Mhz product that is only 3W! This product would have about the same performance as the 5W TMTA product...
There ya go!
EP
Alan -
No changes needed. Intel already sells ultra low power processors into the embedded market. Lower than the low power one you listed. Those same processors would beat TMTA power numbers now if I'm not mistaken. The point is that Intel could go lower right now but so far has not done so.
EP
Alan -
You sound like a typical Intel engineer that is not happy until they have 100% domination. I don't think that is really necessary in this case. The Intel products are very far into the "good enough" category.
Then it should be obvious that Intel could beat TMTA in all catageories if they chose to, including price. They've chosen not to.
EP
Alan -
TMTA Intel
Long battery life 5W 7W
Low cost $85-$198 $96
High performance .5X X
Low power 5W 7W
With a 2x performance advantage Intel still has some room to lower voltage and match or beat that 5W.
EP
Smooth -
Nice to see another knowledgable person posting here. Hope you'll stick around.
EP
Dew -
I think it's important to note that Centrino is not the name for Banias, despite the claim of one of the articles you posted. Centrino refers to the platform of which Pentium M is but one part. Perhaps this was clear to you but maybe not to others. I would also assume that Intel's legal department gave this a thorough going over before blessing it off.
EP
Alan -
Normally that would be entirely possible but would a foundry dick with the receipe? My experience with foundries is that they want to set it and forget it. Nobody gets a special receipe.
EP
Wbmw -
I notice that a die the size of Barton (~10mm/side) would need to have a defect density of 2.0/cm^2 in order to get the yield in that range. Could it possibly be that bad? I would have figured it could be about 0.5/cm^2 for foundries. What's your comment?
I agree. ~0.5/cm2~ would have been my estimate too. The poor yields described most likely include parts scrapped for speed as well as hard defects, unless the rumors Yousef kept referring to about problems with low-K dielectric are as bad as they seemed.... If they are true then you might see AMD changing their design rules and redoing the layout resulting in a larger die. Didn't AMD rework Barfon and the die size went up???
EP
Did these guys copy Yousef's and my posts?
http://www.siliconstrategies.com/story/OEG20030109S0047
SBN Spin: Why did IBM and AMD form an alliance?
Semiconductor Business News
(01/09/03 05:06 p.m. EST)
The New Year has barely started, but there has already been a blockbuster announcement in the industry, when Advanced Micro Devices Inc. on Wednesday formed a technology alliance with IBM Corp.
As reported, AMD and IBM agreed to develop together process technologies for high-performance logic products and microprocessors, at 65- and 45-nm critical dimensions. The first chips built using the technology are expected in 2005 (see Jan. 8, story ).
The deal apparently marks an end to AMD's arrangement with United Microelectronics Corp., a Taiwan-based foundry with which AMD was to develop process technology and build a 300-mm fab in Singapore (see Jan. 8, story ).
So why did AMD leave UMC in the dust and strike a deal with IBM?
There are several theories floating around the market, but perhaps the overriding reason seems clear: troubled AMD needs a stronger partner like IBM to compete against Intel Corp. in the microprocessor business, according to analysts.
Here's another way to put it: AMD's future is doomed--if it continued to work with its now former and fading technology partner in UMC, according to analysts.
AMD has been spilling red ink for several quarters, and, needless to say, is also falling behind Intel in terms of market share in processors, according to analysts. At the same time, the official roll out of AMD's next-generation, 32/64-bit processor line, code-named Hammer, has been delayed.
Fearing that it will fall even further behind Intel, AMD needed a new and stronger partner--and for good reason. “AMD could be dead by year's end,” according to an analyst in Asia, who tracks the semiconductor industry. “The scary thing is that Intel is ramping up two 300-mm fabs in 2003, and there will be lots of room for Intel to lower its CPU ASPs,” the analyst said.
“So, AMD will need the backing of a strong tech partner--IBM--and a loaded financial provider--Chartered/Singapore government--to stay afloat,” the analysts said, referring to IBM's recent deal to forge an alliance with foundry provider Chartered Semiconductor Pte. Ltd. (see Oct. 26, 2002 story ).
What happened to the AMD and UMC alliance? After all, UMC has the capital the fund the development of a 300-mm fab, but AMD reportedly may have been unhappy with the arrangement. “We have heard AMD's CPU yields at UMC are as low as 10-20%,” according to analyst.
Oooo Greg!
I didn't know you had it in you!
EP
Greg -
The CNBC spot with Barrett didn't amount to anything.
Thanks.
Greg -
If you have acces to a TV please let us know if anything significant happens.
EP
Fingolfen -
this poster seriously believes that 1.8 through 2.8GHz are just waiting in the wings????
But they DO believe it. That's why I'm playing AMD shares and options right now. I don't have to believe it, it's enough to know someone else does...
EP
John -
do you think they will be able to pull it off?
I have no way of knowing for certain but I haven't seen the signs yet that one would expect to see if they can do it. So far it's just the old joke about "Big hat, no cattle".
EP
Greg -
I would have no problem with that (or if someone substituted "Intel" for "AMD"). Would you, Ron?
What prompted this post from Ron?
EP
Ron -
Anything? You mean like saying AMD is giving their shareholders a screwing? Something as inocuious as that?
EP