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.
wbmw,
re: Itanium roadmap
Are the performance graphs on slide 7 per core or per processor? It would seem ok if per core, horrible if per processor. (there is an asterisk there with no explanation)
Joe
wbmw,
If you were to take Intel's prediction at face value, they would have forced *themselves* to offer a 64-bit x86 ISA by 2008. The only difference is that AMD forced them to do it 4 years earlier, and the result was a new processor feature that has no applicability to consumers. It's now 2006, and the only segment that can arguably make use of this is in servers and high end workstations. Perhaps Intel should have pulled in the server 64-bit schedule (for Xeon in addition to IPF), but adding it in consumer products has been ineffective up until this point
Good points. Let me add:
64 bit has great applicability to the server market, and AMD matched the timing of the mainstream of the server market nearly perfectly.
One are where I differ - or actually, one thing you are missing - is that in x86 market, moving a 64 bit core (that is already available in the server market) from server market to the consumer market is basically free, because the server and desktop markets use the same core.
Intel's problem with missing the boat, and later being forced to adopt AMD64 are entirely Itanium inflicted. In absence of Itanium, Intel would not have made the mistake of ceding 64 bit leadership to AMD.
Joe
Joey,
LOL. When the sh$t hits the fan, the Droids will be a fun group to watch as their stock plummets.
I will grant you one thing: Intel investors are taking these dog days (more like half of a decade), evaporation of their net worth rather stoically.
Joe
Tenchu,
You can see my point here. The 'Droids aren't upset over bogus marketing deals. They're just upset over those that AMD isn't a part of.
Specifically those that are in violation of the Clayton Act.
It is possible that there will be a day when Intel investors will be upset about Intel's violations of Clayton Act as well. But today, many Intel investors are cheering for Intel's violations of Clayton act.
What I am missing here are level headed Intel investors, who would applaud Intel's achievements on technological level, on merit (if any) and condemn violations of Clayton Act (among others) - and demand that the management stops them immediately. Ongoing violations of Clayton act (see Skype) result in an additional risk to Intel (as an investment).
Joe
ag,
Your assumption is hard to translate to numbers because you need to know how the product mix changes thru the year as the % of dual core parts rises as well as how much 300mm capacity AMD started with as well as how fast more capacity is added. Could you give a guess on the required product mix in Q1/Q2/Q3/Q4?
Here is my last attempt:
I guess the product mix in Q2 may be as follows:
- 70% of 84mm^2 SC, 512K L2 parts for Semprons, A64s
- 5% of 114mm^2 SC, 1MB L2 parts for A64, FX, Opteron (this part will be discontinued under Rev F, IIRC)
- 15% of 150MM^2 DC, 2x512K L2 parts for DC A64s
- 10% of 200mm^2 DC, 2x1MB L2 parts for DC A64, FX, Opteron
http://www.siliconinvestor.com/readmsg.aspx?msgid=22207526&srchtxt=150mm^2
It is possible that the conversion to DC will be faster. But, as far as quidelines for the future, it depends on what Intel does. AMD is not going to push Sempron to DC before 65nm, unless forced by Intel. We may see a start of it in Q4.
Assuming Sempron stays SC, and all Athlons move to DC, I see the situation in Q3, Q4 as follows (ignoting for the moment small die differences of Rev F and possibility of dedicated 256K Sempron die under Ref F)"
- 50% of 84mm^2 SC, 512K L2 parts for Semprons
- 40% of 150MM^2 DC, 2x512K L2 parts for DC A64s
- 10% of 200mm^2 DC, 2x1MB L2 parts for DC A64, FX, Opteron
The above may be the worst case scenario from die size perspective, assuming no SC Athlons, Sempron penetration no higher than 50%, overall implying some traction in the "enterprise".
BTW, the 2 scenarios imply increase in blended average die size from 107 to 122 mm^2, assuming no 65nm.
My guess is that they started Fab 36 at 1250W/W thru Q1 and will increase to 2500W/W by mid Q3 and perhaps will exit the year at 3750W/W.
That seems like an aggressive ramp. My initial thinking was more like starting with some 500 wafer starts per week, ramping to about 1000 by the end of Q1, but given Hector's statements about mature yields, there is not any point in holding back, while there still exists some scarcity for AMD parts in the market place. The only possible complication may be Rev E vs. Rev F issues (if 300mm masks were Rev F, and if Rev F needed a respin).
I would be very aggressive in Q2, ramping to 2000+ wafers starts per week for H2 delivery, then leveling off, based on market demand.
Joe
Tenchu,
Read the press release, Joe. It says "specific Intel processors."
The spin is from the AMDroids, "But, but, but, NO AMD PROCESSORS ARE INCLUDED! WTF!!!" So what? My Northwood processor isn't included, either. Maybe I should sue Intel as well. Too bad I'm not lawsuit-happy like the rest of this country seems to be.
Are you suggesting that in addition to (likely illegal) exclusion of AMD processors (in exchange for money), Skype is also wrong about minimum hardware configuration set for their system?
You can take that argument with Skype, but, IMO, it is not illegal or wrong for a software company to set minimum hardware requirement for their software. It is dumb, IMO, to hardwire it in the EXE, if you asked me, so on this subject, I am with you and I will help you knock down this strawman.
BTW, I have seen a sensible implementation of minimum hardware requirement. In a game Total War: Rome, the system did something to check the hardware requirement of my home computer (that by today's standards is not top notch), and set the maximum resolution to be less than the native resolution of my LCD monitor.
No problem, I dug through config files, found the place, overwrote it, and the program works fine at native LCD resolution (with somewhat lower FPS, which is not a problem for me in that game).
The Skype exclusion is not based on hardware requirements. It is known to everyone who follows computer hardware that the slowest dual core chip on the market is made by Intel. That chip is allowed the max functionality of 10 connections in Skype. ALL of AMD's DC chips on the market exceed that minimum hardware requirement, therefore the restriction is not a hardware requirement.
With that issue out of the way, we are left with the bottom line - Intel paying money to third party to exclude AMD processor. This fits right into the pending AMD lawsuit. What's worse (for Intel) is how blatant this case of quid pro quo is. It is a big mistake by Intel, if you asked me.
And finally, you have repeatedly said that AMD law suit is just another exhibit of law-suit happy US, I guess implying it is frivolous. It is easy to mistake your posts on the subject for saying that all lawsuits are bad. You are not saying that, are you? If so, how can we be a society based on rule of law, if the laws are not applied and adhered to? If a violations, or non-adherence with the law is not brought before court?
Joe
Keith,
1) If the market behaves as expected, AMD should be challenged on the capacity side for most of 2006 (but they´ll of course prioritize the key segments like servers so that constraints there won´t happen)
Given the capacity coming online, that's a very optimistic thing for Hector to say. To reconcile these 2 (capacity growth and still being capacity challenged), Hector must be assuming rapid share gain in Q1 and Q2 (keeping unit sales growing while market is shrinking), then, in Q3 and Q4, at least maintaining share and market growing rapidly (Windows Vista effect).
Joe
Tenchu,
Oh please, will you get over it? If Skype doesn't want to optimize for AMD processors, that's their loss. You 'Droids can go support Skype's competitors.
The answer is in, Intel and Skype are caught red-handed.
http://maxxuss.com/home/skype.html
I am going to lean back and wait the spinners to spin this one. The way I envision this (spinning) is that a person who has no shame (wbmw) will start. Then, people who have no clue (DUrl, smooth2o, tecate) will join in.
Joe
alan,
You got the monopolies backwards...
the difference is microsoft is a monopoly, and as such can not make their product exclusive to somebody. Skype does not have a monopoly on conference calling... or even Internet VOIP.
My point was to demonstrate how anticompetitive Intel tactics is. I think you agree that if Intel struck a deal with Microsoft to exclude or severely restrict AMD processor, that it would drive AMD out of business.
Skype deal with Intel is exactly the same concept, but on smaller scale. If deal with Microsoft (to kill CPU competition) must not be allowed, by the same token, a deal with Skype must not be allowed to stand.
Great move by AMD, BTW, to move quickly and to try to take advantage of Intel's misstep.
Joe
Gee, so Woodcrest may actually beat AMD Socket 1207 to market?
Joe
smooth2o,
Before you call this anticompetitive, why don't you wait for the facts and rulings to come out?
Because this is an investment board. If one does not want to end up with losses (like those of Intel longs), one needs to anticipate the future.
BTW, are you requesting that all speculation stops on this board, or only speculation about Intel's anticompetitive business practices?
IT seems Intel posters are very selective about topics the DON'T want to discuss, limiting the off-limits subject to only Intel's anticompetitive practices. How curious...
Joe
alan,
Given that it was not available to anybody before the Intel purchase, I don't see your case? If it used to be available to AMD customers then you might have something there. It is something nobody had before... now people who buy dual core Intel CPU's can get it for free. what a deal!
Think of a this hypothetical case. Intel pays Microsoft, say, $5 billion to make Windows Vista available only for Intel processors, or limits Windows Vista versions available for AMD processors to Home Basic edition. AMD goes out of business shortly after that.
How is this hypothetical case different from the Skype case?
Joe
wbmw,
Are you sure that's the case?
Based on the available information, yeah. If information supporting the contrary point of view comes out, I may reconsider.
Joe
Alan,
Perhaps that is where we disagree...
I view it as a feature that Intel has purchased (somehow) for use by their customers. If AMD wants to purchase the use of such a feature for their customers, they are welcome to it... they just need to find a partner willing to work with them on it.
If Intel bought anything, it is not a feature. It is disabling of a feature on AMD processor. I am not sure if money changed hands. If so, it would make things even worse for Intel, IMO. Wouldn't that fall under bribery?
Joe
Alan,
If Intel stops marketing because of the lawsuit, then AMD has won without ever being heard in court...
I am not a lawyer, but I would be amazed and astounded if such co-marketing agreements were illegal, even for a monopoly.
Marketing? Intel made a pact with a software company to prevent the software from running properly on competitors CPU. I don't think that is marketing. I think it is anticompetitive behavior at its worst.
Joe
smooth2o,
Can you try a little harder to put some thought to your posts?
Joe
wmbz,
Actually, if intel wrote some optimized libraries that only run that much faster on Intel chips as to allow for 10 streams then Intel could probably license this library to Skype. In their code, Intel could limit the usage of this library to their chips only since those are the chips they optimized for and they cannot guarantee that those optimizations would work on other chips. In that case I would side with Intel.
Sure, Intel's position is likely to be that FUD such as "we cannot guarantee" that some devices or applications planted in the market place to increase monopolization of the CPU business are legal. But AMD position is that only exclusion on merit is legal. If there is no merit to the exclusion of AMD processors, the exclusion is illegal.
Joe
Alan,
sure it does...
No, it doesn't. Specifically, not for Intel.
Intel just handed AMD more ammunition - a mistake, IMO, on part of Intel lawyers. It may turn out to be blanks, but it may turn out to seriously hurt Intel. A no win for Intel from this gamble.
Joe
fpg,
The mobile processor "Merom" will follow with a slight delay in Q4 of this year.
Maybe Intel is planning on testing the new design on the server customers.
Joe
Alan,
That's not to say,
That some INTC engineers did not help them with the code.
Intel could have written the whole thing, but it would still not justify exclusion of competing products.
Joe
Durl,
Well substantiated and thought out analysis. Uh, can I say that on this thread?
Can you do it on any thread?
Joe
chipguy,
What part of
"moving data into the processor more quickly than the front-side bus used by Intel's chips"
I thought you could figure out that if it takes the first processor 50ns to retrieve the data and second processor 75ns, first processor is moving data into the processor more quickly...
Joe
chipguy,
Then write the AMD mouthpiece quoted in that article and
tell him to stop making material misrepresentations of fact.
I was writing to you to point out a slight (less than material) misinterpretation of fact about the speed of Intel's FSB - that speed not being 1066 MHz for issuing of commands, but only 266 MHz.
Joe
chpguy,
Where is this magic DDR memory that "runs at the speed" of a 2+ GHz AMD chip?. Or outruns Intel's 1066 MHz FSB for that
matter.
I thought it was running at 266 MHz. Or. more commonly, 200 MHz and below.
Joe
chipguy,
AMD doesn't have the manufacturing capacity or
design experience to go for monster add on L3s like
Intel uses in its x86 line.
The main thing that AMD was lacking in the past was customers. Now that AMD has customers, sufficient volume of Opteron sales, the biggest hurdle has been overcome.
To maximize performance
with reasonable area constraints AMD should really
redesign a three level cache hiearchy from scratch.
L1 design is tightly integrated within the CPU layout
and pipeline design of any processor. So it doesn't
make sense for AMD to spend the effort at redoing
the cache hierarchy until it redesigns the CPU core
itself.
I think L3 is going to be independent of the individual cores. The cores will just issue a memory request, and it will be served, by the memory controller, either fro L3 or from main memory.
L1 and L2 are closely coupled with the core, so their redesign will come with new core, but L3 does not have the same constraint.
Joe
Keith,
Intel is to add its Enhanced SpeedStep Technology (EIST) to its 65nm dual-core Pentium D 950 processor in order to allow the chip to operate within design guidelines for motherboards, with a maximum power consumption of 95W rather than 130W.
A little weird. Isn't the processor TDP supposed to be independent on whether the demand based switching is turned on or off? It is certainly the case for AMD. More funny TDP numbers from Intel?
Joe
Keith,
If you click on "browse and buy", it says
Our apologies...
It is fixed now.
Joe
Keith,
It looks like the DC 200 and 800 parts will have a 2 speed grade drop in prices. I can't really complain, because, there is likely 1 new speed grade coming out, and initially, when DC Opterons came out, I thought AMD priced DC parts one speed grade too high. So these are the prices I thought AMD should have been charging in the first place.
Joe
chipguy,
You also seem very confused about binning. It costs
the same to make a low bin, mid bin, and high bin
MPU. The difference is what you can sell them for.
That is, if you can sell them at all. If only half bin at the clock speed and power consumption that you can sell, your costs double.
It is highly unlikely that chipsets bin below one bin that Intel sells, It is very likely that a number of CPUs bin either below required clock speed, have higher than desired leakage or both.
Another problem is that if the center of distribution of your bins is too low, yielding next to nothing of the highly desired high end bins, Intel needs to aggressively tweak parameters that in fact produce some of the higly desired bins, but that is done at expense of yields. No such problems exist with chipsets.
Depends on the relative area, process, package, and
test particulars. A 100+ mm2 northbridge chip made
in 130 nm and packaged in a 1000+ I/O BGA might
cost Intel more to make than the 83 mm2, 130 nm
Banias P-M.
Today? Yes. But at the time Banias was the high end state of the art CPU, the costs were much higher. The reasons for that were steep depreciation of the capital expenses, immaturity of the process, the need to bin the CPUs above the threshold that would produce high yields.
It has been a long time since Intel was cruising with greatly binning high yielding Northwood...
Anyway, my comparison was of today's CPUs and today's chipsets, not today's chipset vs. CPU two process nodes back.
Joe
chipguy,
You are wrong. The primary difference between an MPU and
a northbridge is the time and manpower needed to design
it and the methodologies used.
So the fact that the most die of the chipset "bin" at, say 800 MHz, but you sell them at 400 MHz (therefore all are sellable) vs. CPU that bins at 800 MHz (on the same process technology), but you need 4.5 GHz to be competitive plays no role?
Oh yes, and Intel happens to employ an army, and pays 10s of billions to get the CPUs to the 4.5 GHz, and yet it is only spitting them out in low 3 GHz.
If you say that Intel can TODAY produce an 800 MHz Coppermine at the same cost as an average chipset sold today, you would be largely right. But Intel certainly could not do so at the same price 6 years ago. Intel employed army of men counted in 10s of thousands to squeeze the last MHz out of the Coppermine.
Let me get this (since you didn't explicitly go out on the limb with wbmw). Do you agree with wbmw that Intel production cost of either a chipset or a CPU is roughly equal to $30? Something that, BTW, TSMC can do for SiS for some $15, and still make a profit?
Joe
chipguy,
I think you would be very surprised
how large and complex they are. I am talking about millions of
gates of logic plus lots of embedded memory on a >100 mm2,
6+ layers of metal 130 nm device with 1000+ I/Os, many of
them high speed and/or high drive.
Surely they are. But nowhere near the cost and complexity of the mainstream CPUs.
Joe
wbmw,
I claimed that Intel's combined cost of manufacturing a Northbridge and Southbridge is within about $5 of the cost of making a CPU. And you and your napkin calculations could never come up with a decent rebuttal.
I think my rebuttal was to laugh out loud.
Joe
Keith,
re: Dell will not pick up AMD
the outcome may indeed end up being true, I doubt the reasoning expressed in the article is the reason.
Joe
wbmw,
The corollary here is, "I never studied to be a lawyer, and I have no engineering background, but I can do a few calculations on the back of a napkin and tell you all about AMD's and INTC's cost models
Or we can have posts from Salieri of IHub claiming that cost of making a chipset is the same as the cost of making a CPU...
Joe
when equating amd vs intel to david vs goliath, one must wonder how many 'davids' goliath slew before 'the david'
Also, it will be interesting to see all of those 'davids' come back and take a swing a the goliath, when he is no longer the goliath.
Think of the entire Taiwan Inc. for starters...
Joe
sgolds,
I upgraded one server to Windows Server 2003 x64 last week, and it appears that the printers I am using (HP Laserjet 4100 and Color Laserjet 4550) are supported right out of the box. All the other mobo components worked as well. The only unresolved issue is that the ISA Server Firewall Client 64 bit version is not out yet (apparently it is in beta).
Joe
wbmw,
Can you spare this thread of your ongoing non-trades? Anybody who followed your recommendations in the past already lost his shirt...
Joe
sgolds,
I think it is useful to write down the name and number on the jerseys of the companies that are particularly bad about drivers and not to patronize their products in the future.
Canon is off my list for my home printer upgrade...
Joe
AMD CPU and mobo voted to be readers' choice at Tom's:
http://www.tgdaily.com/2006/02/10/readers_choice_awards_2005_winners/page2.html
cruzbay,
The wanna be legal guru posters are total lightweights; we needn't respond to them, and the REAL case will proceed as it will without them. Can we try to adopt a "legal case - don't need to go there" filter on this board?
Sounds like a good idea, but it should be self imposed on part of the posters...
Joe