AMD ingnots, sliced ´TBread´ with the crusts cut off

[jmke]

Some people have questioned the validity of the information in this article

does anyone know what exactly is devianting from the true story?

 

[CTho9305]

Well, for one thing, the article seems to imply that AMD "aims" for different speeds - e.g. some chips they hope come out at 2400+, some they hope come out at 3000+. That would be stupid on their part, since it costs them the same amount to produce a 3000+ and a 1700+. When you fab chips, you "aim" for perfection, and it doesn't happen - you get a normal distribution around some speed grade, and you sell each chip as the one that will make you the most money.

There's been much controversy over Austin's (excellent) Tbred-B guide (from LowYat.net) which thoroughly describes the origin of every series of alphabetic, and numerals in all three lines of code:

"Based on many hands on observations, the 8th and 9th marks (these should be numbers, as shown below) in the second line of the sticker describes the initial speed the processor was designed for (RFBEXR2280073). If the initial model number is higher than the number you have, it means that your processor has been downgraded."


I presume much of the research done for this excellent guide, was extrapolated from information his contact at AMD's Singapore Fab provided. After reading this guide myself, I found every premise empirically verifiable. I've personally examined over 103 JIUHB 1700 TBred-B DLT3C, and 55 DUT3C's (of same). And found 100% of these "lower speed" CPU's were capable of 2100MHz - 2200MHz at default Vcore.

All the cores on a wafer are the same design.

The reason you can overclock is that AMD tests the chips in very hostile environments to make sure that a cheapo computer (crappy power, crappy cooling, etc.) running an AMD processor will be stable even in the Sahara desert in summer. The maximum die temperature is rated at what, 90-95C? Most users keep their processors significantly cooler than this, so they can operate at higher speeds.

edit: A sample size of 100 really is not enough. Also, overclockers rarely test for true stability - many accept a crash once or twice per month and attribute it to something else. Reputable companies can't sell hardware that EVER crashes without help from "acts of god" (cosmic rays, cats, beverages, etc).

 

[jmke]

good info! thanks for your reply

 

[Wingznut]

Originally posted by: jmke
Some people have questioned the validity of the information in this article

does anyone know what exactly is devianting from the true story?

There are so many things wrong with that story that I simply do not have the time (or will) to correct all of it.

But here are a few points:

• If you are asking about the validity that "the 8th and 9th marks in the second line of the sticker describes the initial speed the processor was designed for"... Although I can't speak to the specifics of AMD's methods, I doubt this is true. As CTho stated, it would be silly for them to target certain speeds when manufacturing chips.
• AMD uses 200mm (8") wafers
• The .13µ process uses 248nm lithography, not 157nm as the article stated. And actually... it's been reported that Intel won't even be using 157nm for the .045µ process.
• EUVL is nowhere near ready to use, let alone being currently used on the .13µ process.
• Since the wafer is "stepped" under the photo optics, there is (in theory) no difference between patterning a wafer in the center or the edge. What I mean is... In AMD's case, (most likely) only six die are patterned at a time. The wafer then moves over, and the next six die are exposed under the UV light. This is repeated until the entire wafer is patterned. This is what is known as "stepping."

If you have any further questions, feel free to ask.

 

[jmke]

thanks for your reply Wingznut

There are so many things wrong with that story

we are trying to work away those inaccurate info's & creating an image of the whole that is closer to the truth, your contribution is greatly appreciated!

 

[Liquid3D]

What's sad is that you "don't " or "won't" take the time to "correct" ALL the things wrong in this article. One reason I wrote the article was to coerce accurate information to the surface, not from an egotistical, but pragamatic place.

First: I did not claim were "using 157nm lightwaves" for 130nm gate width's, I wrote "as we surpass 157nm..." The theory of wafer variation is not my own, but Austin's from Lowyat.net. And since his guide containing that info, has been circulating for months, why have you never took the time to criticize IT? Why is it, the "experts" only surface to nullify hypotheses, yet never seem to have the time to educate? Are you not a member of a forum where the exchanmge of information is the primary goal?

Second: I'm familiar with the Intel's decsion to surpass the 157nm process after they were one of it's largest advocates, and I even understand why they've done this. If you can make ad hoc adjustments to "soft lithography" technologies, why invest in an interim stage. It's just too bad they've no concern for the many subcontractors who stand to loose millions. Hey that's capitalism right? To me that's one form of Capitalistic behavior, I choose to label as nefarious (but that's just my theory).
I never claimed my theory was fact, only that it was based on the best information WE "laypeople" have. I don't care if you criticise me, because in your criticism there's tidbits of factual information. Albeit harder to find then the dust microns per meter in a Class 10 clean environment, their there. And on another level the obtuse nature of your emotive response reveals a whole other aspect of the "Fab" world. Which I don't have the "time", nor the "will" to expose. Please forgive my "emotive" defense of the artcile's author, as he did not intend to "incite" any riots. :sun::sun:

 

[jmke]

Hey Liquid3D, I think with the help of the above people we might be able to indeed let "accurate information to the surface"
that way we "laypeople" can get our hands on facts

you forgot a smiley in your post

 

[Liquid3D]

I hear ya jmke it's just I already suspected that chip manufacturers aimed for the highest yeild possible, and then downgraded. And each time I postulated this thoery, I was criticized. So when I write an article which does not in fact expound that theory in the strict sense, but implies AMD either does this, or there could be some anomally which affected their yeild, I'm criticized for that as well. All the while, the only time I hear from "insiders" is when they seem to stop by, give a quick (and usually vaguely condescneding) criticism.

Besides, until an actual AMD fab insider speaks out, how do we know there wasn't some anomally? I beleive I indicated it could have been in the front end, OR back end of the production process. I'm currently finished and editing an article which speaks directly to the other criticism offered up, specifically the 157nm process being eschewed by Intel whilst they make do with current lithography. And I beleive if this person is in the 130nm (.13micron) Fab facility, then he's most likely familiar with 193nm lithography. I knew 248nm was in place until .15 micron, I did not know it was still utilized for .13 micron. For that clarification I thank you. I do appreciate the knoweldge that was shared, I only wish more facts, and less sarcasm were offered.

 

[Wingznut]

My apologies for simply presuming that the authors of this article didn't give a damn about accuracy. I guess I've become a bit cynical after seeing how many people write "factual" articles about this industry, without having much of a concern for what is and isn't fact.

I also didn't realize that it was actually the authors who were looking for the info. I certainly would've given you a better response, had I actually paid attention. So, you have my sincerest apologies for my initial abrupt attitude.

Now that we are past that, let's get to some of the meat of the subject...

• I didn't have "the time to educate", simply because I have a life and a family. A family that would probably ostracize me if I did spend as much time on the PC as I'd truly like. This is also why I will only pick a few points right now.
• I don't believe I have seen LowYat.net's "guide" about his theory of wafer variation, so could you please provide a direct link?
• Constructive Criticism: You might want to re-word the paragraph that starts with "And as we surpass the 157nm process..." I (and probably many others?) read into it that EUVL is being used currently. The whole 157nm part confused me, as well.
• I'll have to disagree with the fact that 157nm tools won't be ready when Intel is ready for them, so Intel has found ways to make due with a previous technology is an example of "nefarious Capitolistic behavior". I guess I'm not sure what you expect Intel to do... Hold off on advancing their own technologies to wait for the vendors to catch up?

I hope I was able to address some of the points, and also was able to offer some constructive information. But I'm going to go take my boys to the park, since it's such a nice day here.

Feel free to post further questions, or look me up on ICQ or AIM (info is in my profile.) I'd be more than happy to help out, as long as the information sought is not confidential in nature.

 

[jmke]

Thanks for your honest reponse Wingznut, the reason I didnt mention in the beginning that the article was hosted @ my site was that I was looking for some feedback concering this whole ordeal, as I know Liquid3D is trying to shed light through some tunnels that have not been explored yet by the "none-insider" people.

When we get to the bottom of this (as in: have all the factual data required for the article) we will update it to reflect the newly found info, thanks again for your constructive responses and hope to hear from you often

edit: LowYat's guide is over here

 

[CTho9305]

Tell LowYat that the die color is the side of the wafer that isn't touched in the fab process, and as such doesn't necessarily tell you anything about the core unless AMD specifically makes "overclocking" cores a different color (which would cost them more money for no gain). IIRC a silicon wafer is naturally black.

 

[Liquid3D]

Originally posted by: Wingznut
My apologies for simply presuming that the authors of this article didn't give a damn about accuracy. I guess I've become a bit cynical after seeing how many people write "factual" articles about this industry, without having much of a concern for what is and isn't fact.

I also didn't realize that it was actually the authors who were looking for the info. I certainly would've given you a better response, had I actually paid attention. So, you have my sincerest apologies for my initial abrupt attitude.

Now that we are past that, let's get to some of the meat of the subject...

• I didn't have "the time to educate", simply because I have a life and a family. A family that would probably ostracize me if I did spend as much time on the PC as I'd truly like. This is also why I will only pick a few points right now.
• I don't believe I have seen LowYat.net's "guide" about his theory of wafer variation, so could you please provide a direct link?
• Constructive Criticism: You might want to re-word the paragraph that starts with "And as we surpass the 157nm process..." I (and probably many others?) read into it that EUVL is being used currently. The whole 157nm part confused me, as well.
• I'll have to disagree with the fact that 157nm tools won't be ready when Intel is ready for them, so Intel has found ways to make due with a previous technology is an example of "nefarious Capitolistic behavior". I guess I'm not sure what you expect Intel to do... Hold off on advancing their own technologies to wait for the vendors to catch up?

I hope I was able to address some of the points, and also was able to offer some constructive information. But I'm going to go take my boys to the park, since it's such a nice day here.

Feel free to post further questions, or look me up on ICQ or AIM (info is in my profile.) I'd be more than happy to help out, as long as the information sought is not confidential in nature.

You have certainly laid to rest my pre-judgemental "reaction" to your initial criticisms. I'm disabled, and therefore, I'm in a unique position to spend many hours on-line, where you have responsibilities, which are more pressing then answering what in fact may be "wild" assumptions about the industry. It's true I did not have nearly as much information as I would like prior to writing such an article. It's frustrating having a passion for computer science, and even a greater passion for the truth, yet to be surrounded by so many conflicting theories, vehemently defended based on ego. In so far as my mention of the 157nm ultrviolet light-wavelength, "being surpassed" this is semantically ambiguous, and on the surface misleading. I apologize for being defensive. Much of my frustration derives from the misinformation which is so prevelant in so many forums, and on that point we obviously agree. And i do understand why it's not always worth your time, nor your responsibility to correct every inaccuracy in every forum. If I worked in the industry, I doubt very much I'd come home, spend hours searching the forums for misinformation, then arugue (in the negative sense of the term) with hard-headed individuals about that which I knew to be true, while I neglect my family.

Just so you know my heart and mind is in the right place. I obtained the gist of my information from publications such as EETimes, BYTE, Datamation, Fabtech, Semiconductor International and Kuro5hin, to name a few. 90% of the motivation for this article was in fact based on my prima facie belief in Austin's "TBred Guide" here's a link; http://lowyat.isentral.com/guides/athlonxp/
This guide has been reprinted in so many hardware sites, I can't begin to name them all. And this is not to defer any criticism to him!!! I extrapolated from, and promulgated his theories emneshed with my own. And I not only do I accept full responsibility for my not investigating further specific facts, but do not presume to be "mislead". I chose to base many of my premises from information contained in that guide, and if it's inaccurate I still give the authors credit, because I do not belive their intentions were anything but honorable. I would not presume to be some victim of of misinformation.

In so far as the overlcokability of AMD's Tbred-B cores, I've always believed that one yeild, no more, is the intent in wafer fabrication. In fact I've put forth that in several "article" length posts I've published at Xtrmemsys. And in so far as misinformation is concerned I've written about certain contradictions in logic concrening Tbred's and the experts artciles about them. For example; http://www.xtremesystems.org/forums/showthread.php?s=&threadid=8249

In so far as my inaccuracies pertaining to fab specifics. I understand the cost of a one photomask containing a specific level of the chip's architecture, costs approximately $1-million to produce. With 16 - 24 masks per core architecture, it would be economical suicide trying to fabricate several "speeds" from the same design. And even the assumption this would be achieved using mutliple templates per each PR or speed on second thought, is a risky postulate. I also found the following quote from my article is contradictory in this respect, and inaccurate in it's description of fab costs;
"The photomask or stencil (which becomes the template for the design) costs approximately $1 million to $6 miilion to produce.
EDIT: Thank you for enlightening me, as to the actual cost for the "entire set" of masks totals $1 million.

As I am starved for information, and a philosopher at heart, my intention in publishing this article was simply to understand how the fabrication process "really" worked. I wanted to explain why AMD Thoroughbred-B0's were overclocking with the headroom they were, and in the absence of all the facts, I hypothesised. (making the infamous Inductive Leap) Additionally my inspiration was my spending $200 for an AIUCB 2400 which cannot reach beyond 2400MHz at any Vcore, and then purchasing a $60 JIUHB 1700 which reaches beyond 2480MHz at 1.8Vcore? Perhaps you can see the juxtiposition in logic therein. Not withstanding the information, from "experts" such as Ed Stroglio who published the artcile; "How to tell them apart" (Ed Stroglio Overclockers.com) which makes the follwing claim;
"You can identify which type of TBredB it is by looking at code that begins the second line of coding on the processor. If you see a code like "AIUHB" that begins with the letter "A," that's a high-end TBredB.
If you see a code like "JIUCB" that begins with the letter "J," that's a low-end TBredB"
http://www.overclockers.com/tips00173/
Perhaps you can see why I become frustrated. From the data I've seen, JIUHB has been the highest clocking Tbred yet? I wrote to Ed about this months ago.

Whilst so much misinformation is circulated YOU are perhaps the first person to step foward willing to provide factual "inside" data. For this I'm grateful. Again I profusely apologize for my terse retort. PLEASE feel free to email me after you have time to look over Austins "TBred guide." And perhaps Ed Stroglio's article as well. You'll find I'm willing retract and clarify any contradictions, or inaccuracies in my arcticle. Please accept my apologies Wingznut as my reaction, was simply due to MY frustration simliar to yours, concerning inaccuracies. I do not wish to propagate any misnomers. I only desire to educate myself, and in my frustration attempting to find information as elusive as IC design specifics, I prematurly judged your critque. I would be honored if you would take the time to email me, and point me in the right direction.

 

[pm]

Like Wingz, I have a few comments on the article and like Wingz I don't have time today to actually write them up. My family is always interfering with my plans for living in front of my computer.

I'll type up my non-company specific commentary and email it to you by Thursday night.



Patrick Mahoney
Microprocessor Design Engineer
Intel Corp.

 

[Chaotic42]

Originally posted by: pm
Like Wingz, I have a few comments on the article and like Wingz I don't have time today to actually write them up. My family is always interfering with my plans for living in front of my computer.

I'll type up my non-company specific commentary and email it to you by Thursday night.



Patrick Mahoney
Microprocessor Design Engineer
Intel Corp.

If it's not asking too much, would you mind posting it here? I'm personally very interested in the subject.

 

[Lynx516]

Just noticed some other iregularities I would like to correct in that article. Though I am glad to see people trying to educate people on this subject.

"These 2 digits are the laser marker ID and used for chip tracing according to a friend who works in AMD Singapore Fabs. But we don?t quite understand the actual meaning of ?chip tracing? as from what we know, the earlier production of Athlon like Thunderbird, Palomino do not have these extra 2 digits...

Maybe those CPU with one particular code for example 24 was originally designated to be made for 2400+ model.
But due to failure of achieving that speed, it will be remarked to lower model. So whether that CPU can hit 2400+ or not with default voltage doesn't help...because it has a great chance that won't run at that speed with default voltage at all..
We should treat these 2 digits as for reference only, because not 100% of the 28 chips can run at 2800+ !!
Bear in mind that the maturity and purity of wafer also affect a lot in performance. Coding will be the 2nd consideration after confirming which core is better by your eyes... o your 6th sense..
Anyway to be safe, you should pick one Thoroughbred with 25, 26, 27 code instead of 24 and 28 because we have no idea which is really the best one. Judging by our experiences after reviewing more than 50pcs Thoroughbred B 1700+ with different batch and production timeframe, code 25, 26, and 27 have less ?disappointment?."

This is a load of rubish. It is very unlikley that the digits mean anythign about what speed they are intended to run. They are most likly for tracking chips which die under unusual curcumstances. And I wouldnt be surprised if on RMA many resellers have to report the numbers inculding this one to AMD iin order to detect manufacturing flaws.

ALso what most people dont realise is that the "stepping" code on the die is just realy a manufacturing prosess revision number which will inherently determine how well it OCs. They are not intentionally done for low end and high end tbreds.

Another point:

"The third line of code known as the stepping specifies the manufacturing process in reference to the wafer. These four letters are the four first letters of the TBred-B's; A > J > K > R. A being the highest quality, and indicative of those cores closest to the wafer center. A being the purist, and therefore capable of the highest speeds. So there are A TBreds and J Tbreds. However by AMDBoard article "There have not be any AIUHB Revision B's, so the next highest grade is the JIUHB", therefore all are JxxxB 1700+ Tbred-B's."

This isnt true. The whole wafter is used but due to the probablility of errors in the waffer increases the further out you go you are more likly toget dead chips not ones which work but dont OC so far.

"During fabrication, any contaminants that land on a wafer -- the sheet of silicon that's the foundation for a chip -- can ruin the processors built on it. For example, a dust mote glowing in a beam of sunlight can damage hundreds or thousands of circuits. Even a smoke particle, measuring in at 0.5 microns, can short out a pair of lines in a nascent processor."
src: Byte.

Hence why they use clean rooms so their is no dust or smoke. The things that muck up the slilcon is strucural defects in the wafer

 

[Sunner]

Originally posted by: Chaotic42

Originally posted by: pm
Like Wingz, I have a few comments on the article and like Wingz I don't have time today to actually write them up. My family is always interfering with my plans for living in front of my computer.

I'll type up my non-company specific commentary and email it to you by Thursday night.



Patrick Mahoney
Microprocessor Design Engineer
Intel Corp.

If it's not asking too much, would you mind posting it here? I'm personally very interested in the subject.

Ditto that

Im just a lowly sysadmin, but still, I love reading about this stuff as a hobby

 

[jmke]

Liquid3D has wrote an addendum to his original article, hope this helps clearify certain points!

Over here

 

[Shalmanese]

Does anybody else find it funny that a couple of Intel engineers are giving "advice" on a expose about AMD?

Just a couple of notes: If you still can, make the original version of the article also availible. You look more transperant and less like you have something to hide if you keep all revisions of the article.

If you have deleted the original file, keep a note of this for future articles.

Dont post in bold... Just dont. Its hard to read.

This is not meant to be offensive in any way but your writing style is tedious to read and gets in the way of your message. I know the internet is not exactly the gold watermark of journalistic standards but improving your sentence structuring and the "flow" of your article would go a long way.

 

[jmke]

Originally posted by: Shalmanese
Just a couple of notes: If you still can, make the original version of the article also availible. You look more transperant and less like you have something to hide if you keep all revisions of the article..

the original is still available! the link posted is page 4 of the article, the previous 3 are from the orgininal one!

 

[sgtroyer]

Okay, a few points.

One of your big questions seems to be why can I overclock a 1700 by 100%, but I can't do the same with a 2400. To me, this doesn't seem all that surprising. They're the same processor, right? Same design, same masks, same process. Some just come out faster and others slower. This isn't due to mistakes or impurities, it's normal. There is always random variation in the fab process, and some wafers come out fast and some slow. AMD tests the finished wafers to see how fast they end up. Now, if a certain fab is 'fast' for a month or two, there may not be any slow parts, but they still have to sell 1700's. So they just label them 1700, even though they may actually qualify for a higher speed. Other 1700's get the label because that's all they can do. There's no conspiracy, it's just AMD's way of selling all of the processors they manufacture, and getting more money for the good ones. If you get lucky, you'll buy a 1700 that's actually fast enough to be a 2400. You probably will not, however, buy a 2400 that is actually fast enough to be a '3300'. It's already towards the higher end of the speed distribution, so there isn't much left to gain by overclocking.

A lot of what you refer to as 'insider' information is actually publically available, you just have to look for it. I think many of the factual errors you made were not confidential information at all: the information has been published. I'm not saying it's always easy to find information, but you can't just throw up your hands and despair of understanding the process because you're not an insider. Many good books have been written that explain silicon fabrication techniques, and journal articles go into great detail about modern methods. It's work to wade through the literature, but if you're going to present yourself as someone knowlegeable about fabrication, then it rests upon you to make sure your facts are straight. When you publish an article, people trust you to have checked your facts and assumptions. By presenting what you imagine to be true as actual truth, you are misleading your readers and doing them a disservice. I hope you don't feel I'm attacking you, that's not my purpose. You're curious and want to know more, like many of us, and that's great. I'm by no means an expert either, it's very difficult to become one. When you publish an article, though, you're effectively saying "I know what I'm talking about." It's not acceptable to do your fact-checking after the article is published, by that time the damage is done.

A few more minor points:

You state that the price for one mask is $1 million. Actually, for .13um, it's about $1 million for the entire set of masks. One individual mask is ~ $30,000 or so.

The fact that Intel processors can be clocked at >3GHz while AMD hits the ceiling at considerably less than that has nothing to do with a conspiracy on AMD's part to keep you from overclocking, or anything else mysterious. The processors are fundamentally different designs. The P4 has a longer pipeline, so it does less per clock cycle, roughly speaking. This lets them clock the chips faster. Athlons just can't be clocked as fast as P4's.

There certainly are 'physical' limitations to making faster processors, but I wouldn't say there are 'metaphysical' limitations. Metaphysics is the branch of philosophy that explores the nature of reality. So I guess an AMD engineer could question whether her experience of processor design is reality, or just a dream, but I doubt that's what you meant.

Actually many of us think processor design is a dream job, but that's another matter altogether.

As always, take any of this information with a grain of salt. I think I'm correct in my assertions, but I may not be, and appreciate corrections.

Regards.

 

[pm]

Nice post, sgtroyer. Very good points.

Actually many of us think processor design is a dream job, but that's another matter altogether.

If you have an inclination towards engineering... yes, I think that's definitely true.

 

[jmke]

great info steve! thanks for your reply

 

[Harabecw]

silly question that should not be in here. yet: where do you even start to get that kind of a job? I'm at the right age (18-19) and have a job because I'm saving for university.
so...years before I'll really achieve what I want...where is the start?
consider that I live in israel...

 

[sgtroyer]

Where do you start? Well, get yourself to a good university, and get a degree in electrical engineering (or computer engineering, but I'm partial to my own specialty). Do really, really well, and get lucky when you're done. It's hard, but not impossible. Intel hires a lot of engineers, although some jobs are more interesting than others, and the really interesting ones are pretty hard to get. Israel is actually not a bad place to be. Intel has a design group there that did the Pentium M (Centrino) processor for laptops.

For some jobs advanced study (Masters, PhD) is requisite, but not for everything.

Anyone else want to weigh in on this? There should be a category for this type of discussion.

 

[Harabecw]

Already to planning on some form of engineering. perhaps computer engineering, but I'll have to see when I get there.
I'm also wondering about how much I should save. probably enough not to worry about the first 2 years.