AMD ingnots, sliced "TBread" with the crusts cut off
The following is an addendum, and amendment of my prior error laden articles on the same subject. Please remember this is merely my hypothesis. I originally posted this at Madshrimps, but thought my "Home away from Home, away from Home" should have it too>
Recently AMD enthusiasts (overclockers) have had a smile across their faces the length of which must have wifes and girlfriends wondering. Their new-found mistress; "lower-speed" Thoroughbred-B's, and their "double overclock" potential. Almost mythical in overclocking circles (prior to phase-change cooling) doubling a processor's default speed is the apotheosis for the enthusiast. So where did these sweethearts come from? And why have I chose the title "AMD ingots sliced "TBread" with the crusts cut off"? Buckle up, it's going to a bumby ride!
For those of you not familiar with microprocessor fabrication I'll give a breif description, and you'll see it's pertinent. The silicon wafer, from which processor cores are cut, begins life as "...a purified polycrystalline silicon...first melted, then encouraged to crystallize into long cylinders, called ingots." This is actually a growth process taking a "seed crystal" which is submersed in a bath of molten silicon, and when the seed is slowly pulled up, the Ingot forms. These ingots, are then sliced into "wafers" 300mm or 12" diameter. Next the wafer is polished on one side to a mirror perfect surface, there cannot be a flaw in this surface. Next a photo-resist or photo sensitive chemicals cover the wafer which is imprinted with the microcicuirty pattern as ultrviolet light shown through a stencil (or photo-mask 16 to 24 mask layers are required) leaves the image behind. "...the remaining photoresist is removed, leaving a pattern of silicon dioxide on the silicon wafer. Additional materials, such as polysilicon, which conducts electricity, are deposited on the wafer through additional lithography and etching steps. Each layer of material has a unique pattern. Together, they will form the chip's circuitry in a three-dimensional structure." The photomask or stencil (which becomes the template for the design) costs approximately $1 million to $6 miilion to produce. Lithogrphic equipment itself weighs upwards of 20 tons, and cost $17 million. The cost to construct and operate a .13 micron "Fab" plant, runs approximately $1.6 billion. The costs are very high, and these figures pertain to a fab plant built to accomodate just one core design. For this reason, it is often the case mutiple processor core speeds are produced exploiting the design parameters until the fastest possible speeds are reached given design's architecture. it's not that AMD doesn't WANT to make faster CPU's it's just that their are metaphysical limitations, given the current fabrication technology.
And as we surpass the 157nm process (.13 micron) traditional photolithographic methods are reaching their limits. When using a .13 micron chip, this is the measurement across etched lines in core architecture along which electrons travel (bytes/bits). The smaller the gate width, faster the processor speed and lower voltages. Using a process known as EUVL (Extreme Ultraviolet Lithography) a process by which the light waves which were once used to imprint down to .13 microns across, can now be shortened to ten times as small. And simply stated, processors will soon be ten times as fast! It's estimated by 2007 there will be 30GHz processors in your PC. EUVL uses lightwaves so short lenses are now obsolete, in gfact 90% of the light would be absorbed unless the process takes place in a vaccum, and utilizes coated mirros to bend or focus image into the wafer. This is not science fiction, in 2001 at Scandia national Labrotories, a prototype EUVL machine succesfuly demonstrated the technology. Recently Lawrence Livermore National Laboratory, Lawrence Berkeley National Laboratory and Sandia National Laboratories collaborating as the Virtual National Laboratory, successfully transferred the EUVL technology under a multi-year CRADA (Cooperative Research and Development Agreement) to the Extreme Ultraviolet Limited Liability Company (EUV LLC), a consortium headed by Intel Corporation that includes chipmakers Advanced Micro Devices, IBM, Infineon, Micron Technologies and Motorola. (Tech Space Feb 12, 2003) There's even a more promising technology LADI (Laser Assisted Direct Imprinting) utlizing a excimer laser, developed at Princeton which I won't go into, suffice it to say it has an even greater potential then EUVL. And it circumvents the 300 or so steps which make Ltihographic, wafer fabrication exuberantly costly.
The reason I've explainied microchip fabrication, in such detail, and with an emphasis on cost, is to exemplify the propensity, for error, and the ad hoc cost control.
There's been much controversy over Austin's Tbred-B guide (from LowYat.net) wich thoroughly describes the origin of every series of alohabetic, 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 pressume 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. For this reason, I feel the guide's predictions are astute. In fact I surmise the author of bloodys.com has written his [COLO=blue]Athlon XP/Thunderbird Processor Identifier[/color] based upon Austin's theories. You can demo the program HERE. Most likely after a barrage of criticism Austin amended hie guide. to include the following;
"Update:
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”."
It could be the case there's semantic obscurities in the translation, as I happen to believe the 8th & 9th digits hold pertinent data. In fact in the photo at the bottom of this artcicle, (the sticker from a Barton 2500) reads 25 in Code line 2, 8th & 9th places. I currently awaiting a reply from a Barton 2500 owner with 27 in these places. I predicyt, it will be an excellent overclocker.
Finally pertaining to the article's title; Tbred slices with the crust cut off. I want thank antipop at Xtremesys for his scrutinty. He described the ID process as "nonsense", and this is where being open minded is key. Instead of vehemently defending a moot point, I decided to investigate further, based his next (criticism) question;
"One last question, it's about the place on the wafer, you say that A and J are the closest to the center of the wafer, but i've never seen a Tbred B mark with a K or an R, what do they do with the chip on the border? Do they just put them in the bin?"
In this question lay the answer to the conundrum. Yes, I beleive do believe AMD "binned" the cores from those area's of the Wafer. Due to a possible lense misalignment, or some anomaly during the multiple Photolithographic/etching cycles, the outer wafer cores were rendered unusable. This is the section of Auistin's guide to which antipop was referring;
"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 would unambiguously explicate the JIUHB 1700's (and 2100's) propensity for 40% to 105% overclockability. I purport an anomally in either the photolithographic lense alignment, or some other manufacturing anomally rendered K - R outer areas of the wafer inutile. The high number of variables during manufacture, which can have detrimental effects on the end-product are as plentiful as dust particles in the air;
"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." (BYTE)
This would ceratinly explain why the lower speed TBred's are capable of double overclocks, while the highest PR's aren't doing much better then 2.5GHz to 2.6GHz this under phase-change temps.. Ceratinly AMD utlizes their best cores for the upper end models. This theory would also explain WHY the small percentage of lower speed TBred's capable of 2.5GHz to 2.7GHz overclock's, commensurtates perfectly with those of their counterparts (models 2400 and above). If AMD's Fab facilities are producing cores capable of such extraordinary overclock performnace amoung the the lower speed models, why aren't models 2400 and higher reaching 3.0GHz and above? Certainly the DUVL process hasn't been completely exhausted for AMD, ergo a ceiling has been reached. The fact Intel is surpassing 3.06GHz using standard Deep Ultraviolet Lithography, implies the problem with AMD lie somewhere else. It would be naive to think AMD wouldn't want their PR's 2400 and above capable of 2.7GHz overclocks, because this would simply be the basis for 3600 or 4000 PR processors. Even the experts were befuddled with the Tbred-B. I knew something was amiss when Ed Stroglio (Overclockers.com) began identifying these as "Low end / High end" Thoroughbreds. It seems even Austin subscribed to A - J vs K - R as a means of determining performance;
"From our personal findings, the purity of the centre wafer will determine the high end model of a CPU. Higher end models are then called an A Thoroughbred while the lower end one is called the J Thoroughbred. AMD even had produced R Thoroughbred and K Thoroughbred for Revision A. But so far they do not exist in Revision B."
All one need do is empirically verify the data for them-self. Either recording overclocks from signatures, or revisitng as many threads on the topic (overclocking 1700's, or 2100's)
as they care to examine. Irregardless, I do not think it coincidental the highest overclock acheived by the fastest Thoroughbred, is equal to the highest overclock acheieved by the slowest Thoroughbred. There's something missing, and that is those outer wafer cores. Ergo the "crusts cut off the Tbread". pardon the pun.
I want to thank everyone whom contributed, and especially thank Austin @ LowYat.net. Without such experts (enthusiasts) willing to take risks sharing their hypotheses, this would be a boring hobby.
Very good article, seems to be a lotta people studying into these 1700/1800 Tbred B chips atm, should be some good reports coming out in the next few months about them hopefully.
__________________
"There is only one thing better than having a linebacker in your office, and thats having two linebackers in your office"
12.7Ghz folding for XtremeSystems
Re: AMD ingnots, sliced "TBread" with the crusts cut off
Quote:
Originally posted by Liquid3D
Finally pertaining to the article's title; Tbred slices with the crust cut off. I want thank antipop at Xtremesys for his scrutinty.
Wow i'm famous
Your theory is really interesting and that might explain why amd isn't able to reach intel speed. 0.13µm desing haven't reach its limit yet.
I was akso thinking that the letter A J K R would have someting to do with the purity of the wafer, and that amd has been able to make better wafer and so that the K and R aren't needed anymore (one can remember how the Tbred A were poor oc'er)
The fact that intel is faster (i'm talking only in MHz) lie also on the internal design, IIRC intel has longer pipeline than amd, and that would make some calculation slower but you can benefit from the higher clock speed that the design allows you.
Also i wouldn't be surprise that amd does throw quite a few wafers due to some misalignement or some dust. Maybe they have a problem with one factory and that would cost higher to repair it than to use it as it is. All this of course are simple hypothesis.
Anyway kudos liquid3d for your research and your openess
__________________
The page you are looking for is currently unavailable. The Web site might be experiencing technical difficulties, or you may need to adjust your browser settings.
Thanks people coming from the members here that's really a compliment. I consider you guys some of the most astute enthusiasts on the entire web, and that's no "blue smoke" being phase changed in reverse, if you get my meaning.
__________________
I hate to advocate drugs, alcohol, violence, or insanity to anyone, but they've always worked for me.
Hunter S. Thompson (1939 - 2005)
Very interisting writings, Liquid3D.
Although I must admit it was a bit hard for a french like me to read easely such a technical post, I have to say that it's kinda amazing to see how numbers are speaking.
Don't get me wrong, but oc'ing can be quite predictable with such bases, not to an extent of precision, but quite accuratly. I've seen a lot of message where peoples ask "how high will this chip oc etc etc (quite typical, naye?)", and if the replies once were only "check yourself!", now based on such numbers, we can at least (for the sake of knowing ) predict it a bit, for what it worth.
It's kinda like decoding the Rosette Stone you have made, Liquid3D.
__________________
The page you are looking for is currently unavailable. The Web site might be experiencing technical difficulties, or you may need to adjust your browser settings.
Please, not the mouse-cord! It's means that I consider you, along with others here, "senior" members. Your opinion matters, like; TheDude, antipop, Method Man, jmke, gobbo, sysfailur, MrIcee, Tweaked, mdzcpa, Morphling1, saaya, OPPAINTER, JCVigg...etc. People I've learned from, and whom I believe to be knowledgeable.
I'm just a guy who likes to tinkle with all sorts of things, the past year I haven't had the time to play around with any of the cool hardware released, have a prometeia & watercooling + 172watt pelt lying next to me here, but simply dont have the time to use it
I don't think I mentioned this is the article, however; unless there's evidence either way, what's to say AMD's "Laser ID" system was intended for the end-user? being an internal coding system, I don't beleive we can dismiss the possibility they either intended to relase a 2500 TBred, or are simply using the digits to "track" the chip's PR under certain conditions. Perhaps the third alphabetic, in code line 3, proceeding the date may have something to do with this. M indicating a "combined" batch which didn't pass QC tests for any number of reasons, may have some reciprocatory relationship with those numbers?
And I beleive people are mislead by the semantics of Quality Control. Because a chip didn't "pass" QC doesn't necessarily mean it can't attain a high clock speed, even at default Vcore. Perhaps the QC issues in this case, pertains to the absence of the outer wafer core steppings, albeit by manufacturing error or otherwise. Forced into relabling chips intended for high PR some marking would have to indicate where the chip originated from, irregrdless of the circumstnaces. Ergo, what constitutes a QC failure may have nothing to do with a performance "failure" per se, but simply a necessity to re-route the piece?
Originally posted by Liquid3D being an internal coding system, I don't beleive we can dismiss the possibility they either intended to relase a 2500 TBred, or ...
Originally posted by Liquid3D Please, not the mouse-cord! It's means that I consider you, along with others here, "senior" members. Your opinion matters, like; TheDude, antipop, Method Man, jmke, gobbo, sysfailur, MrIcee, Tweaked, mdzcpa, Morphling1, saaya, OPPAINTER, JCVigg...etc. People I've learned from, and whom I believe to be knowledgeable.
BTW thanx all.
Thanks, you are on my list of senior members, Id even say you know more than me!
thanks to the Airplay that this article got , we're getting different reactions from all over, this one is translated from Dutch, it was posted on a very popular HW site www.tweakers.net in the Netherlands
Quote:
"The whole story of those rings is crap, your idea of the CPU production is just wrong. If the chips in the middle are better then those on the edge, it's is a clear sign that your manufacturing process is wrong!
*RobT is chip designer and gets quite some info on the manufacturing.
That with the pieces of dust and bad lenses means that they don’t know anything about it; the wafers concern only a few chips (so, the negative of 2*2chips or so), so the chips on the edge are lighted and produced in the same way as those on the middle. So if there’s something wrong with the lenses that light it or the wafers in the whole construction, then e.g. the chip located right under the wafer would be transferred poorly on the wafer. Then you’ll finally get a wafer with a chessboard pattern of bad chips
Dust? Sorry, but those chips are made in a class 10 environment, if it isn’t already a class 1 by now (the number represents the amount of dust particles that’s bigger then 1micron, for every cubic metre); you can have a few bad chips on a wafer due to such impurities, but half of them??In that last case I’d want less people smoking in the waferstepper. …
If there might be something wrong with the silicium, crystal or impuritiets, then you’re not going to throw away half of the chips but instead have a talk with you supplier.
There are much more stupidities in that article.
The reason why the lowest clocked CPU’s (overclocked) obtain the same speed as the highest clocked, means that the manufacturing process is okay, meaning it’s got a high yield. That means (what’s already been discussed a lot) that all CPU’s obtain the highest speeds (it’s more or less limited by design or thermic reasons). All CPU’s are about the same quality, obtain the same speed, so AMD probably rates them by power usage, heat dissipation etc.
Don’t forget that these overclockers don’t do the same tests as AMD, or actually: not at all.Remember the PIII 1.13, they were called back to the factory when THG discovered they weren’t able to do Linux kernel compilation but the CPU’s passed the Intel tests anyway.
In that same way do the OC’ed T-Breds pass the overclocker’s test, but AMD found out that e.g. they absolutely weren’t stable enough or so. Or not 100% stable with all PSU’s, coolers, mainboards or so."
There are a couple of things I'm going to have to agree with in the criticism based on what little I know about semiconductor companies (btw, I work for one of the largest, although not in the fab).
First, the idea that a semiconductor company like AMD would let some sort of problem in its process that causes it to throw away large percentages of chips (i.e., a substantial yield problem), is in my experience, wrong. The first thing they would be doing is figuring out what the problem was and correcting it in a HURRY. Bad yield numbers mean higher costs, and nobody is willing or can afford to put up with that these days.
Second, with respect to dust and other contaminants - as said above, these chips are made in cleans rooms so clean you wouldn't believe it. Again, if there was a problem like that, AMD would be fixing it in a hurry for the same reason they would fix any other yield problem - to keep costs as low as possible.
Third, the reason higher rated chips can't hit much/any higher than the best lower rated chips is architectural. AMD has hit the limits of the t-bred/barton architecture. Intel can hit higher speeds with basically the same manufacturing process because they designed PIVs differently, with longer pipelines, etc., so they would be able to push the clockspeeds to higher levels.
My guess (and it is a guess) is that AMD has their process working really well and the 1700+ chips are coming from the places on the wafer that they know from testing aren't quite as good as the spots the higher rated chips are coming from - they're just not much different because AMD has their process working so well.
Anyway, the post is a nice effort, but I think its based on some incorrect assumptions.
I actually have been expounding that theory all along. I know that AMD produces one core, and simply lables down from there. I think because this was challanged so many times I simply began to question it myself. The center wafer theory, is not my own, and i did indicate that in the article. i also proclaimed Austin's guide was "excellent" and perhaps I'd done better to write from the basis of it's validity or inaccuracy. However I tried to imply in the absence of inside factual info, we enthusaists are often left to theorize.
In so far as AMD controlling manufacturing standards to such a dgre, I admit I never knew to what degre that was, however; it's only logical they would try to correct any anomalies, as the must conctinually check the peices for any anomalies. Now I know there's much more control then i'd previously assumed.
In so far as air quality, i definately know the least acceptable level is still 1000 times cleaner then an operating room. And I read the following quote fom an article before writing the article;
"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."
I wrote the article to coerce crtiism, and bring facts to the surface. I'm not offended by criticism, and especially when presented in the manner whihc you have. I respect and appreciate your not taking the opportunity to be condescending, or insulting. Thank you
It's means that I consider you, along with others here, "senior" members. Your opinion matters, like; TheDude, antipop, Method Man, jmke, gobbo, sysfailur, MrIcee, Tweaked, mdzcpa, Morphling1, saaya, OPPAINTER, JCVigg...etc. People I've learned from, and whom I believe to be knowledgeable.
Linear Mode
