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.
Spokeshave -
I must admit it is a beautiful piece but couldn't you have gotten a cheaper one at K-Mart?
Spokeshave -
I am going to improve my marketing and communications efforts so that agenda-driven FUDsters have less fodder.
You do that. Meanwhile your competition will be improving their silicon.
Haddock -
I have to respin the A64 layout anyway to get rid of some unexpected speed paths so I can get over 2GHz.
OK, now I'm your boss:
After all this time I would think that you are nearing the end of the line for speedups from layout changes and many of those are done in metal anyway. You need to fix the process and that may not be possible this generation. Bigger L2 means new base layers, not just metal. What new speed paths are you going to introduce in addition to the ones you are trying to fix? You could have been shipping A64 ever since April if you just used the Opteron die. Why didn't you do that while you were doing your redesign? A 6 month slip costs us dearly in MSS not to mention reputation. How many units will you have to ship just to pay for the design effort and where is the crossover point compared to an A64 version of Opteron? All this is assuming you won't introduce any new problems and you won't slip your schedule any more than you already have. Why don't you just ship an A64/Opteron and put your efforts into 90nm?
Spokeshave -
Yes, yes. So you keep pointing out. What you are missing is that most of the design work for Hammer was done when AMD had plenty of each.
No need to be condescending. It takes all the fun out of discussing things with you.
Given what you said above, you need to come to grips with the fact that AMD expected the 256K version to be competitive and they expected SOI to be manufacturable at a much earlier stage. Those things didn't happen the way they planned it. Time to implement plan B, or more like plan X/Y/Z by now.
So you Mr. Spokeshave are now the Manager of this project. Your product is too slow to compete, the process yields poorly and you need a much larger die to hope to reach decent ASPs. Your company is hemorrhaging red ink at a record rate and you now need a new unforcasted design team to redo the layout and cache clocking so you can add some more L2. You'll have to spin a few mask sets at about a million a pop, and your validation lab will have to make room for a complete new A64 validation, all over again when you change the cache and layout. As if they didn't have enough to do trying to validate Opteron at 2/4/8 way. At the same time you're under pressure to prepare your product for an eventual 90nm transition but you don't have Design resources to do both much less the money. There's a 1Meg Opteron that's already validated and you could just package it differently and call it an A64. No redesign needed, no validation needed, no muss no fuss and you weren't going to ship high volume anyway! Now your Designers can roll off onto 90nm where the real action is going to be anyway.
What are you going to do Boss? It's your call.
Haddock -
By gross size I mean the size you get by just cutting off the bits you don't need, so I don't think it's up for discussion unless i goofed with my ruler. The result is not rectangular, so you probably need to do more than that (but the result does fit in a rectangle that is smaller than the Opteron rectangle).
I understand what you meant and I don't disagree with you on that point. I just don't think they went to the trouble of redoing it. Not that they couldn't, just that they didn't.
Haddock -
The gross size of A64 looks to be 167mm2. About 100mm2 of that is cache.
Maybe, maybe not. I wouldn't argue for a second that the A64 die couldn't be reduced to that size. It certainly could be. My point is that it probably isn't because of time, money and resources. All of which are in short supply at AMD.
Time will tell.
Keith -
No, it´s not.
That was a joke Keith.
Keith -
No offense intended but you are rambling and making no sense.
Let's call it a night.
Keith -
If I could figure out what you just said I'd be happy to respond.
ChipGuy -
Stop lying. Everyone knows Intel has to *give* them away, unlike Opterons which sell like hotcakes through major OEMs.
It's a miracle my post wasn't deleted...
ChipGuy -
My opinion is that if you are going to claim some knowledge above the pedestrian level you should be willing to stick your neck out and give some perspective. If you're wrong once in a while then well, life goes on. At least you start people thinking.
Keith -
I´m pretty sure that they were confident to reach these speeds, unfortunately, the well known problems have caused the shortfall. This certainly doesn´t qualify as an intended lie. They fell short of their target because of unexpected technical(as they openly admitted recently) problems. I think that describes it better, and explains why you´re the only one accusing them of "lying". I understand your dissapointment, I´m sure everyone else feels the same way. The SPEC numbers show that K8 performs just like they´ve promised, actually a little better even
My intent was not to accuse AMD of lying, but to point out that Intel wouldn't have been given the benefit of the doubt, which you are generously willing to do in the case of AMD. When AMD made those claims, they could not have had any certainty that they would be able to deliver what they claimed, yet that didn't stop them from making such claims. What word would you use to describe a statement made that someone has no certainty they can fullfill? Would you use the same word if the company's name was Intel? Somehow I don't think so.
You believe AMD made that claim in good faith. I believe you can not make such assumptions about a company losing money at an unprecidented pace and desperate to keep perspective customers interested.
ChipGuy -
So you will know who you are dealing with.
"Andy Grave" was a regular poster on SI under the alias "The Watson Youth". He claims to be a retired Process Engineer from IBM. I have no reason to doubt him. He claims considerable expertise, which he may have but he is very reluctant to share any of it, as you have probably noticed.
Does anyone know how many processors AMD sold last quarter?
AMD has stopped announcing their processor sales.
Windsock -
Continued losses and continued cash decrease will result in AMD's inability to pay its bills when they come due. That is the recipe for bankruptcy, or if you prefer BANKRUPTCY.
You've got to understand. The important thing is how cheap and affordable AMD products are and how well they do on benchmarks. Please don't bother us with unimportant irrelevant matters that don't matter.
Well apparrently 1500 Itanium2s sold in a single system!
HP dominates Top500 computer list
http://www.infoworld.com/article/03/06/20/HNhptop500_1.html
Probably the biggest winner on this latest list is Intel Corp., which jumped from having 54 systems in the November rankings to 119 in June. A 1,500-processor Itanium 2 machine at the Pacific Northwest National Laboratory gave Intel's 64-bit processor its first top-ten showing on the list.
Well apparrently 1500 Itanium2s sold in a single system!
HP dominates Top500 computer list
http://www.infoworld.com/article/03/06/20/HNhptop500_1.html
Probably the biggest winner on this latest list is Intel Corp., which jumped from having 54 systems in the November rankings to 119 in June. A 1,500-processor Itanium 2 machine at the Pacific Northwest National Laboratory gave Intel's 64-bit processor its first top-ten showing on the list.
Spokeshave -
OK. Sure. And you base that assumption on what? You say that you are "sure". What information do you have that you base this confidence on?
Relax. Take it easy. I was just agreeing with your assertion for argument's sake.
You want to believe that the 1Meg L2 A64 is a different die from Opteron then fine. Neither you or I will change a thing. Either it is or it isn't.
There were similar hints that Opteron's performance would be poor as well, but it has performed quite admirably, and has for the most part exceeded expectations.
Not entirely in my book. AMD promised it would beat anything Intel had to offer in SPECint and SPECfp. They didn't. If Intel had made the same claims and fallen just as short then people here would be saying that Intel had lied. None of the AMD fans said AMD lied, why not? There as some very impressive scores for 2 and 4-way configurations but some or even most of those numbers may fall soon to shipping Intel systems well before any Opterons are offered by any known brand name. It's a good start, that's why I'm an AMD investor, but AMD can't keep out of bankruptcy Court by posting benchmarks on their website. As an Investor, I care about products and they have to do more than wow the gamers who blast mutant aliens. They have to sell and at a profit. That's where AMD still has a long way to go.
Spokeshave -
That makes no sense at all. If they have a different die for the 256K version, why would they not use the same die and just add more cache for the 1M version? It would not seem to make any sense to needlessly manufacture an A64 with a disabled extra memory channel, and two disabled extra HT links, especially when there is another die design that does not have these parts. I would imagine that AMD will do everything possible to make the A64 die as small as possible to increase the number of die candidates per wafer.
I'm sure it was AMD's plan to only use the 256K version and that's certainly a different, much smaller die, but when it because apparrent that the performance was just too poor to be competitive they were in a real pickle. What were they going to do? Redesign the 256K version? You can't just stack more L2 on there without redoing the clocking and layout. Mask sets are approaching $1million a set. Did they have an extra design team just sitting around doing nothing? Would they want to turn a few mask sets when they're already posting multi-hundred $$million$$ losses quarter after quarter? There's a 1Meg L2 Opteron sitting there that you can put in a different package and establish a market presence as an A64 while you're ironing out the process. The volumes won't be much until 90nm. Why spend all the money on something that's low volume and has to be redesigned for 90nm soon anyway?
That's my opinion. Let's wait for facts to see who's right.
So, what do you estimate the yield to be?
Spokeshave -
Now that you no longer have the power to delete embarrassing posts I have taken you off ignore.
Where are you getting 193mm^2? I have not seen a die size for A64 published yet, and it seems that you are assuming that A64 will be the same size as Opteron. That is not a fair assumption.
I am assuming that Opteron and A64 are the same die. I don't know that for a fact but it seems most likely to me. As they both have the same 1Meg of L2 it would make more sense to just use the same die. The cost of maintaining 2 designs and doing 2 sets of bug fixes, speedpath fixes, transistor revs etc would be too much for a company on a limited budget. With the low volumes expected early on it just wouldn't make sense. The 256K version though probably is a different die.
Could I be wrong? Sure, but that's my guess.
TWY -
I get about 140 whole die assuming 16% kerf and edge loss. What are your assumptions? It appears you are assumming almost 30% kerf and edge loss. That's absurd.
Absurd? I don't think so. Try this:
http://www.icknowledge.com/misc_technology/die_calculator.xls
You don't even get 140 gross die with 0 edge die exclusion.
TWY -
At 193mm2 there would be about 120 whole die on a wafer:
What is the justification for this??
You're not serious are you? You can't even figure this one out? You claim to be a professional. I'm totally amazed.
I tell you what, you are way in the hole on your question/answer ratio. In fact you have provided zero answers. You start answering some questions for a change and I'll answer one of your's.
TWY -
Somehow... I just knew you wouldn't provide a number. But, in a subsequent post, you claimed a direct cost of $100 for A64. Whether you realize it or not, that implies a yield (using reasonable # for processed wafer cost) of about 20% Is that what you are claiming for current yields of A64? If not, then what are you claiming for current yields?? And....please don't tell me that a $100 direct cost doesn't imply a certain yield level.
At 193mm2 there would be about 120 whole die on a wafer. A 20% yield would give 24 GDPW. A $100 die cost would give a processed wafer cost of $2400. I think that's too low for SOI, BWDIK?
Again I ask, how about you providing a yield estimate? You do a lot of criticizing the estimates presented by others. How about sticking your neck out and offering some estimates of your own?
Semi -
And I'm "sure" that the same AMD fans who crucified intel management for reissuing stock options to employees that were under water, and screwing the non-employee shareholders, will now all chime in with the same exact criticism's of AMD..... Right?
You're not holding your breath are? We'd hate to lose you...
Another example of why Windows isn't ready for prime time as a workstation OS.
I run VNC from home over my VPN to access my workstation and save myself the time of going in. No problem with Unix. With a highspeed cable connection it's almost exactly like being there, just a little slower. I also access my PC in the same way but I find out that when I turn off the screensaver it turns it off at work as well! The machine is completely open to anyone. Windows has no concept of a virtual terminal and no ability to allow multiple users to have graphical access to a single machine.
Just another example of how far superior Unix/Linux is to Windows as an industrial strength OS.
Joe -
Your [wbmw] post implied that you would give us a break with your QS spam until Athlon64 benchmarks are available.
A64 benchmarks are already available and they aren't good...
http://www.tweakers.net/nieuws/24744
http://www.xbitlabs.com/articles/cpu/display/athlon64_7.html
http://www.theinquirer.net/?article=8313
http://www.x86-secret.com/articles/cpu/k8-1/k8-8.htm
Gateway to intro Xeon Servers, may introduce an Itanium Server:
http://www.idg.net/ic_1322257_9675_1-5124.html
CeBIT: Gateway plans new handhelds, server
Tom Krazit, IDG News Service\Boston Bureau
June 19, 2003, 09:00
More details about Gateway Inc.'s forthcoming PDA (personal digital assistant) and a new server to be released in August were revealed by executives Tuesday in a phone interview at CeBIT America in New York.
The Gateway 100X will be Gateway's first handheld when it is released on Monday with the launch of Microsoft Corp.'s Pocket PC 2003 operating system, said Mike Stinson, vice president and general manager of mobile products for Gateway. The company considers its PDA a high-end product that it will sell for between $300 and $350, he said.
The Poway, California, company will include both a Compact Flash and a SD (secure digital) slot in the PDA, Stinson said. It will come with 32M bytes of RAM and 32M bytes of ROM, he said. Information about how much RAM would be user accessible was not immediately available.
Gateway will use a 3.5-inch reflective TFT (thin film transistor) display on the Gateway 100X, with a screen resolution of 320 pixels by 240 pixels, Stinson said. The PDA will also come with a 400MHz XScale processor from Intel Corp., he said.
The 100X is only the first in a series of PDAs and smartphones, Stinson said. Gateway will roll out other models over the coming months, at different price points and with different features, he said.
Also among the 50 new products Gateway plans to release this year will be a four-way, rack-mounted server based on Xeon MP processors in late August, said Scott Weinbrandt, general manager for the systems and networking group. The company will also "investigate" the possibility of releasing a server based on Intel's Itanium processor, he said.
Gateway will also add two storage products to its lineup by the end of the year, one in the third quarter and the other toward the end of the year, Weinbrandt said. The company plans to focus on direct-attached storage products, he said.
Keith -
More importantly, afaik, Durons were still being PRODUCED last quarter.
I thought Durons were produced in F25 exclusively and that's now 100% converted to Flash.
?????
Jun 19, 2003 (FIND, Inc. via COMTEX) --
Pursuant to section 221 of the Trade Act of 1974, an investigation was initiated on May 14, 2003 in response to a worker petition which was filed by a company official on behalf of workers at Intel Corporation, Systems Manufacturing Technology Development, Hillsboro, Oregon (TA-W-51,749) and Intel Corporation, Systems Manufacturing Technology Development, DuPont, Washington (TA-W-51,749A).
The petitioner has requested that the petition be withdrawn. Consequently, further investigation in this case would serve no purpose, and the investigation has been terminated.
Signed in Washington, DC, this 6th day of June, 2003.
Linda G. Poole,
Certifying Officer, Division of Trade Adjustment Assistance.
Haddock -
Elmer, do you have me on ignore? If you don't reply, I'll understand .
No. Should I?
Keith -
why don´t you just say what you´re implying?
Ok, I will. Assuming shipments start in August, AMD doesn't know at this time what their binsplits will be on A64 with any real confidence. They think they can deliver 1.8GHz in enough quantity for HP to prepare a system for Sept availability. Anything above 1.8GHz would be welcome but unknown at this time, however AMD would prefer to leave the impression that 1.8GHz is not necessarily the top.
As for 90nm, yes I did say that we should expect to see A64 pushed out to 90nm and I still think that's true, at least for any meaningful volume with a 1Meg L2 die. What could change that? If AMD could contract out Athlon to a foundry they could free up enough capacity to make a few more A64. That would drop performance on Athlon because the foundries won't process Athlon to the limit like AMD does (not a knock here, Intel does the same). Performance drops and ASPs do the same. AMD could even lose money on the contracted parts so I don't expect this to happen.
I'm only guessing here but I don't think A64 and Opteron will be different die if the A64 really has 1Meg of L2. Too many designs to manage and mask sets are much too expensive to keep flipping steppings for 2 cores at the same time, 3 if there is a small die 256K L2. Intel wouldn't do it and they have money to burn.
ChipGuy -
Factor in higher raw and processed wafer cost, higher scrappage, lower device yield, and higher packaging and test costs for A64 and it is easy to see for a wide range of direct vs indirect cost breakdown for Athlon leads to the conclusion that it would cost AMD at least $100 to produce a 1 MB A64,
perhaps much more.
Why would these guys care? The most important topic of the day is how well it will do in DoomIII. Nobody cares about making a profit here.
Sgolds -
Damage control from that HPQ pdf specification, which only showed the 1.8GHz 3100+ system. AMD wants the world to know that this is just one of the processors being released officially on September 30, that production is running smoothly and that both varients will be in the hands of system builders in August.
I would only change one word in your post.
AMD wants the world to think that this is just one of the processors being released officially on September 30.
YB -
you are known to be good in estimating yields and costs. If you tell me A64 costs $100 to make I will trust you.
I don't claim that. I don't know what AMD's yields are on SOI, yet.
wbmw -
Why don't you tell me why that is instead of lecturing me about forum netiquette..
It's called denial. If they could acknowledge it then it wouldn't be denial.
ChrisC_R
Well, that's about all the FUD I can take for tonight.
Fine. I'm polite and try and help you understand. Your response is to insult me. You are now a member of my ignore list.
ChrisC_R
Oh, so your analysis only considers simple wafer defect density - defects in the silicon crystal lattice. I thought your analysis addressed process defects too.
Wrong again. That's 2 for 2. In this context defect density means defects introduced during or as a result of the manufacturing process.
But WTFDIK......
I'm not going to touch this one......
ChrisC_R
Didn't I see you already do that, and come up with a cost of about $25 per chip? Even that estimate used a defect density such that more than half the silicon would be thrown away.
Nope. I showed how it would yield on bulk silicon, not SOI. And as for the yield loss, that's what happens when you have a big die. Use the same defect density with a smaller die and you'll see the yield goes up in terms of percentage.
How so you know that AMD won't get much higher yields? They have been working on it, you know:
I suppose it's always possible for AMD to all of a sudden take a problematic process technology (SOI) and have far and away better yields than anyone else in the world does on a much less troublesome technology, bulk SI, but I wouldn't bet on it.
"Advanced Micro Devices Inc. (AMD) has developed a technology called Advanced Process Control (APC) that allows chipmakers to make fine adjustments to their tool sets while wafers go through the manufacturing process. AMD claims its homegrown technology is unmatched."
Please... You want me to start listing all the things AMD claimed that didn't come true?
Maybe there are many people in Dresden who can get the defect density down:
Maybe. Let's hope so.
Do you really think that the best yields AMD can expect are less than 50%?
The bigger the die the lower the yield in percentage. It's not an AMD thing. If you had 100 die on a wafer and 50 defects you'd see about 50% yield. If you had 200 die on the same wafer and the same 50 defects you'd see about 75% yield. Same process, same wafer, same defect density, different yield percentage. It's the big A64 die that makes the yield so low. A64 needs a huge cache to compete with Intel and this is the price they pay, so to speak. Big die, low yields and restricted capacity.
DARBES -
I suspect that some adult diapers need changing over at inteL.
For a 193mm2 SOI processor that only makes it to 3100+ and maybe will showup in Sept?
I don't think so...