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.
Commrad Billski!
You will find this a boring thread. Come over to the INTC thread..
EP
Bitter little man -
There goes the neiborhood...
Unfortunatelly you don't understand PCI-Express, or anything else for that matter. Try considering 8b10b.
Now if you can behave yourself I'll allow you to address me. Don't blow your chance now.
EP
John -
I have no clue, how much value does Solaris, and Sun support, have?
I don't really know but it's the only card SUN has to play.
EP
John -
Two desperate companies. Would the whole be bigger than the sum of the parts (I may not have that quote right)?
The question is, would SUN have AMD? SUN needs someone and Itanium would do as well as Hammer. SUN is selling Solaris or they're just another vendor with nothing special to offer. Is Hammer a better choice than Itanium to bet the farm on? SPARC has no future and SUN certainly doesn't need to spend the R&D on it. They can't afford it. Let someone else develop the hardware. Can SUN trust AMD to deliver? I don't think so but who knows?
EP
wbmw -
I have a little AMD news for the thread.
The confusion seems to involve the 2400+ as well. The page on pricewatch shows that vendors are listing the XP2400+ as both 1.93GHz and 2.0GHz. Even they don't know what it is...
EP
John -
Looks like the MSFT lawyers will have a really good weekend celebration.
Looks like everybody will have a pretty good Monday. Markets are up pretty strongly in after hours.
Here's an idea for you. What about an AMD/SUN merger? Both companies are in the toilet. SUN needs a competitive processor and AMD needs a big name server win.
EP
John -
I think it's only a matter of time before SUN dumps SPARC. The question is what they will replace it with? Itanium? Hammer? Xeon? My guess is that SUN will have to make a decision before Hammer proves itself to be manufacturable and able to compete on performance. I see SUN going Intel but for some reason they don't always take my advice...
EP
Intel Says It Won German Lawsuit Against Via Technologies
By Alan Patterson
Hong Kong, Oct. 31 (Bloomberg) -- Intel Corp., the world's largest chipmaker, said it won a lawsuit in Germany against Taiwan's Via Technologies Inc.
``The Dusseldorf District Court has decided to prohibit the importation or sale of Via chipsets that work with Intel Pentium IV processors,'' said Josie Taylor, an Intel spokeswoman in Hong Kong.
Via spokesman Richard Brown and spokeswoman Gaynor de Wit didn't immediately respond to calls from Bloomberg News.
Via is the world's second-largest maker of personal computer chipsets, which work with processors to manage functions such as memory and graphics display.
wbmw -
The PCI-Express clock runs at 250MHz. A single link, as you say, can deliver 2.5Gbps (or 2Gbps of raw data). However, PCI-Express is scalable to multiple links. If Intel wants to have an advantage over the current AGP 8x standard, they will need to use a x16 PCI-Express connection for 32Gbps, or 4GB/s of bandwidth
Exactly. Or even more connections.
EP
wbmw -
AGP 8x has 2GB/s of bandwidth, so the article is either wrong about the bandwidth of PCI-Express, or the doubling with respect to AGP 8x
I believe PCI-Express is 2.5Gbps which is 8b/10b encoded so the actual bandwidth is 2Gbps.
EP
New PCI standard soon to replace AGP
United States--The arrival of the Intel-proposed AGP port was meant to complement PCI by being the dedicated interface for graphics acceleration. However, the AGP will come to its bottleneck with the arrival of the 8X graphics acceleration chips, and it is possible that the 8X AGP will be its last incarnation.
As higher bandwidth requirements emerge for graphics applications, a new chip interface to replace the AGP port will be needed. The most probable solution is a PCI Express interface (formerly 3GIO or third generation I/O). The PCI Express can deliver bandwidths of up to 8Gbps per channel, about twice the bandwidth for an 8X AGP Pro interface. Another advantage of the PCI Express is that it is scalable for more channels and higher frequencies when required to address applications that demand higher bandwidths.
The PCI Special Interest Group, headed by top computer firms like IBM, Intel and HP have been defining new I/O interface standards that meet the growing bandwidth requirements of the world.
Rich Baek, executive director of the PCI SIG, revealed that PCI Express and PCI-X 2.0 will be the two mainstream I/O interfaces for PC products by 2005. The PCI-X 2.0 can support technologies for server environments like 2 and 4 Gigabit Fiber Channel, Ultra640 SCSI and 10 Gigabit Ethernet. According to Baek, products supporting PCI Express, like mobile and desktop computers, and products based on PCI-X 2.0, like high-end desktops and servers, will emerge in 2003.
Taiwan has about 90 member companies in the PCI SIG, the third largest delegation of members after the United States and Japan. Members pay $3,000 a year to avail of firsthand technical information and the right to vote on a proposed specification. Baek has encouraged more (Taiwan) companies to join the SIG and obtain updated information for PCI. "We hold free compliance workshops and developer's conferences around the world for our members. Taiwan has been one of the most important developing centers."
wbmw -
We'll have to see how it turns out, but I had no idea that a foundry's generalized process was so low in performance.
That's what Yousef and I have been laughing about. AMD wants everyone to believe that UMC has some superprocess. Couldn't be farther from the truth.
EP
wbmw -
As far as our previous discussion on die costs, our aim was to figure out the total cost of manufacturing for a processor, not just the silicon costs. These costs would include test and packaging, and any other relevant expenses.
I'll guess $0.50 for test and a buck or two for the package. Anyway $5 would still leave you enough change for a big mac.
Earlier, a few posts had argued that Intel's Banias, for example, would lose money if it were sold for less than $100. It would be nice to have better data to refute those claims than just silicon costs.
Backend isn't anywhere's near the expense of silicon. If it costs $500Million to develop a processor and Intel sells 100 million of them then it's $5 per processor.
EP
Windsock -
I've done some research and confirmed your statement that depreciation is included in wafer cost.
This site shows a breakdown of wafer cost which includes depreciation and all sustaining costs. It puts a 300mm wafer cost at $3367.25, close enough to the numbers we've been using and it includes all expenses except R&D.
http://www.icknowledge.com/economics/300mm_costs.html
This link put's Intel's defect densities at between .2 and .3.
http://www.icknowledge.com/economics/yields.html
"The resulting defect densities for a large die (Pentium 4 size of 146mm2 is assumed) are approximately 0.2 to 0.3 for the IDMs and 1 to 2 for the foundries."
Taking the 2 extremes shows that the yield for a die the size of NorthWood could vary by a factor of ~6 between a top running fab and a foundry. AMD is counting on a foundry for their future products.
EP
Windsock:
The wafer cost does include depreciation and is a large component of the ~$2500 for a 300mm wafer.
Then where am I going wrong here? If we assume that a fab is depreciated over 3 years, and assume 7500 wafer starts per week, that comes out to over $1700 per wafer for depreciation alone. So $800 covers the cost of the raw wafer and labor as well? Is 3 years too short a time for depreciation?
EP
Some thoughts on die cost:
Numerous estimates have Prescott's die size between 80-90mm2. If we just use 100mm2 to be conservative and a $3000 300mm wafer cost, again more conservative than the $2500 mentioned by Winsock yesterday, we get a die cost of a little above $6. To be less conservative and use the aggressive numbers the die cost goes down below $4.00. Of course R&D and Depreciation must be factored in but the raw die cost is incredibly low.
Here's another article that confirms the die cost numbers I posted yesterday.
http://zdnet.com.com/2100-1103-915032.html
"It probably costs Intel $21 or less to pop out the silicon inside a Pentium 4 on its most advanced manufacturing lines"
Spokeshave -
your estimate of $25 of less is probably a pretty good one, but the number is meaningless without R&D and depreciation.
I wish we really understood depreciation. Over how many years is it written off? If we knew that we could get a pretty good number per wafer with some basic assumptions about fab loading. Add that to the yielded die cost and you're close to the mark. R&D expense is going to be a function of volume.
EP
Winsock -
The cost of a 300 mm wafer is about $2500.
Thanks for reminding me. I completely forgot that Banias will almost certainly be on 12" wafers and that lowers the die cost considerably.
EP
Spokeshave -
Either you have intimate knowledge of the costs associated with a product that is not yet even in production, or you do not understand what the word "speculating" means. I will go out on a limb and assume that you are neither prescient nor privy to Intel's cost structure that even the most informed analysts are not privy to. Therefore, it would seem impossible for you to have cost information on an unreleased product. I must then assume that you do not know what the word "speculating" means. The conclusion is inescapable. Is that a fair assumption?
No it's not, for several reasons but I'll just give you one. All we need to know is die size to do a pretty good job of estimating cost. The original poster excluded R&D in his estimate so I did the same. I also showed how his estimate of depreciation was not realistic.
Why don't you do what nobody else but me seems to be willing to do. Estimate product costs based on yields and wafer cost. What's the problem here? This isn't rocket science. An 8" wafer costs about $2000 or less and that pays for the people who do the processing. You can get a pretty good idea of yields if you know the die size.
http://www.icknowledge.com/misc_technology/die_calculator.xls
Use .25/cm2 as a good estimate of world class yields.
Add a couple of bucks for packaging and test.
Like I said, what's the problem here?
EP
Spokeshave -
Nonetheless, all of this is pure speculation, and we both could be 'way off.
You're the only one speculating.
EP
DewDiligence -
EM: are you including the cost of owning and running the fabrication plant that is used to make the chip, including the salaries and benefits of the personnel at the plant, electricity and other utilities, depreciation of the equipment and real estate, etc.?
To some extent I am. Non depreciation costs are included in the $2000 wafer cost I use plus some depreciation, but let's take a closed look at depreciation and assume none is included in the number I used. Yor numbers are about $75 higher than the ones I come up with. So at $75 a die, how many die would be necessary to pay off a fab and how long would that take? @$1.5Billion for a modern fab that would be about 20 million banias processors or roughly 6 months output for a small fab if we use the number of 150 GDPW. So your numbers have a fab fully depreciated in 6 months. All I can say is I don't think so.
I’m glad you posted because now I can see why you and wbmw are way too low on your estimates for Banias’ COGS and consequently way too high on your estimates for gross margin. Regards, Dew
I'm glad you posted too because it gives me a chance to dispell some of the misinformation posted here.
EP
DewDiligence -
Your contention that the $637 high-end Banias will have a COGS of only $60 – and hence a gross margin of 90.6% (577/637) -- strikes me as naïve. Regards, Dew
I've got to jump in here. The fact is that the $60 cost is unrealistically high unless you are amortizing a heap of R&D. Based on die size the cost will be well below $30 and probably below $25.
EP
DewDiligebce -
Please note that these figures do not even include any allocation for R&D costs – just the actual cost of goods shown on the P&L statement. The arithmetic yields a cost-of-goods range from $105 for the low-end version of Banias to $204 for the high-end version. Hence my original comment that INTC cannot sell Banias at a profit for less than $100. Regards, Dew
We welcome other points of view here but you're way off base on this one. By excluding R&D the cost of Banias will probably be well below $25, regardless of speed. It's easy to calculate based on wafer cost and die size.
EP
Intel's processor sales in China surge 39%
By Jack Robertson
EBN
(10/25/02 01:29 p.m. EST)
Beijing -- Intel Corp.'s PC processor sales in China jumped 39% during the third quarter compared to the same period last year, according to the firm's top China executive.
Wee Theng Tan, president of Intel China Ltd., said the company's strong showing compares with market analysts' estimates of a 20% year-over-year growth rate in the country's overall PC market.
Tan said all market segments -- distributor channel, Chinese OEMs and multinational OEMs -- are evenly divided in Intel's processor shipments. "We see no slowdown in the China market," he added.
One factor spurring China's PC bonanza is the surge of sales in outlying regions of the country. "PC sales in these regions, where there has been a low penetration up to now, are growing even faster than in more mature markets of the major cities," according to Tan.
The Intel official said Chinese consumers are buying high performance processors as soon as they come on the market, following the same sales pattern as the rest of the world. He said the new 2.6-GHz and 2.8-GHz Pentium 4 chips are selling in China at the same rate as in the rest of the world.
Intel's new 64-bit Itanium 2 processor has also been selected by several major Chinese server firms, including Legend Computers and Langchao, he said. "Itanium will find its way into the Chinese marketplace."
The ability of Intel to ship Itanium 2 chips at many performance levels to China is due to a recent relaxation of U.S. export control limits on computers. "We are not meeting any (export control) obstacles today on all our commercially viable processors," Tam asserted.
Intel's own chip production in China at its assembly and test plant in Shanghai is expanding capacity in a $300 million project, on top of the $200 million original investment in the facility. The chip plant assembles and tests more than half of all Intel-produced chipsets and half of all flash memory devices, he said. Early next year the Shanghai facility will start assembling mainstream Pentium 4 processors.
Unlike many foreign electronic firms, Intel sees no need to form manufacturing joint ventures with Chinese partners. "We have been operating well in the current environment on our own," Tan pointed out.
Intel Capital, however, has invested in more than 20 Chinese companies, mainly software developers, systems integrators, and wireless product firms. He expected some Chinese firms will be selected as part of Intel's announced $150 million investment fund for companies in the wireless market.
DewDiligence -
On the possibility that Banias, in any flavor, can be sold for less than $100 at a profit, we'll have to agree to disagree. Even after reading your analysis of Banias' architecture on the TMTA MB, I simply can't come up with a set of plausible assumptions that leads to a profitable price below $100.
Lets see your numbers. I'd like to see what your assumptions are.
EP
wbmw -
Isn't R&D expensed immediately? I don't think it's amortized like depreciation would be. This is an area that's a little unclear.
EP
Bob -
All of them that I've seen have been enormous. Have to mount perpendicular to the motherboard in a slot just like any card.
Those aren't P4s. They're Xeons. I thought you said the P4 were huge.
The problem is that there are no multi-CPU P4 motherboards out there (that I know of) and that if there were, they'd be huge because of the size of that CPU.
The P4 will not run in dual or multi processor mode. You need an Xeon for that.
EP
Bob -
The problem is that there are no multi-CPU P4 motherboards out there (that I know of) and that if there were, they'd be huge because of the size of that CPU.
Bob the CPU is not huge. In fact it's quite small.
EP
wbmw -
If Prescott ends up outperforming Hammer, and I think there is a definite probability for that, then AMD's higher costs are going to put them in a tough position to compete with price.
Maybe AMD can't get SOI to yield? That would kill any desktop hopes.
Do you remember the Prescott die size? Seems to me it's really small.
EP
Saturn V
I was intrigued by the statement that Manitoba is already sampling. Does anybody know if Manitoba uses 0.13micron or 0.09 micron lithography ?
Prescott will be the lead vehicle on 90nm and I believe Intel will be using a different 90nm process alltogether for their comms products.
EP
Greg -
Uses the term "HT II", which I'm guessing is Hypertransport II. Something big is coming in the second half.
HyperThreading
Re: New Intel Factory To Open
Now all Intel needs is customers.
wbmw -
I expect Intel to follow them soon with a unidirectional source synchronous protocol of their own.
Do you see PCI-Xpress playing a role here?
EP
Tim -
You have made the claim that I have repeatedly accused you of being a drug addict and being fired from Intel.
I have explained to you several times that I did not make that claim but perhaps for some reason your recollection is faulty. I will repeat. I have no knowledge that you are on drugs, nor do I have direct knowledge that you were fired from Intel. What I did say, or imply was that drug use would explain your bizarre behavior and basic bitterness toward mankind. Same for being fired from Intel. In short, you're a real sicko, in my opinion.
So now we get to statements made by you. You claim that I have repeatedly accused you of these things. That statement is false and if intentional could be considered libelous in itself. On what do you base this? I can understand that you may have misunderstood my original post, which I have clarified for you, but you claim that I have "repeatedly" made this accusation. Where are the repeated posts? Please provide proof that I have repeatedly accused you of in fact using drugs or in fact being fired from Intel. 2 posts will do.
EP
Test Post
Tim -
Rather than annoy people here I have taken this discussion to the "Parking Lot".
We can continue it there and anyone who gives a crap can follow.
EP
Heidegger -
I'm waiting for your proof that I am either on drugs or was fired by Intel. You made the claim. Let's see your proof.
This is getting a bit old. I already answered this question a couple of posts ago, perhaps for some reason your memory isn't working correctly and you forgot? I did not say you were on drugs, I merely offered it as a possible explanation for your bizarre behavior, nor did I state as a fact that you were fired from Intel, I simply offered it as a possible motive for your bitterness toward life. You keep making this claim but you are incorrect. Besides, I have clarified the statement so even if it wasn't clear to you the first time it should be now.
EP
It seems to me if the people on this thread want it resolved on this thread then what's the problem? Is this thread here for the members or the administrators?
EP
Heidegger -
You have claimed that I am posting on drugs. You have claimed that I was fired from Intel. Provide support for these claims, or retract them. And if you persist in repeating them, IHub faces legal responsibility.
You are mistaken. I did not say those things. I asked if they were the case or suggested them as possible explanations. I did not state them as facts.
I think you have a mistaken idea of what constitutes libel. You didn't know the difference between libel and slander to begin with. First one can not libel a fictitious character posting under an alias. Who knows who you are? Second there is no claim of fact, only opinion and no matter how reasonable that opinion is in this case, there are other explanations involving other pathologies. Perhaps the first explanation is the preferred one because it's the easiest to correct?
Why not get back to your insightful posts and stop all this?
EP
Heidegger -
The IHub censors are behaving as we have seen so many times in so many fora, both on the Net and in the real world. Once they start deleting and rewriting history, they find they have to delete references to the deleted history, lest the sheeple get confused.
That's why I have requested a clear statement of the rules because the original criteria was no SPAM, profanity or personal attacks. Those rules have been expanded in some undefined way. I would like a definition.
Worst of all, they sometimes delete your libels, sometimes they don't. And they usually delete my defenses against your lies.
I have clarified that my statements were theories to account for your bizarre posts. They aren't lies, they do a pretty good job of explaining the facts. If you post your defense in a manner that doesn't violate the rules then in my opinion your post should be allowed to stand.
EP