AMD warns on third-quarter results

Late yesterday, AMD warned investors that its results for this year’s third quarter will be lower than expected. The company previously said revenue would climb 8-12% from the second quarter to the third, and it forecast a third-quarter margin of "approximately 47%." Now, the company expects a 4-6% revenue increase and gross margin of 44-45%.

Interestingly, AMD lays the blame almost squarely on GlobalFoundries for the missed targets. Take a look:

The less-than-forecasted preliminary third quarter 2011 revenue results are primarily due to 32 nanometer (nm) yield, ramp and manufacturing issues at GLOBALFOUNDRIES in its Dresden, Germany factory that limited supply of "Llano". Additionally, 45nm supply was less than expected due to complexities related to the use of common tools across both technology nodes. AMD continues to work closely with its key partner GLOBALFOUNDRIES to improve 32nm yield performance in order to satisfy strong demand for AMD products.
The less-than-forecasted preliminary third quarter 2011 gross margin results are primarily due to less-than-expected supply of "Llano" and associated products with higher average selling price (ASP). Additionally, shipments of AMD’s next-generation server processor, codenamed "Interlagos", occurred later in the third quarter than originally anticipated.

In other words, hiccups with the 32-nm ramp at GlobalFoundries purportedly prevented AMD from meeting demand for both Llano chips and earlier, 45-nm products. Oh, and Interlagos shipped too late—but that’s no great revelation.

AMD says it will reveal its full third-quarter financials on October 27 after the market closes.

Comments closed
    • sschaem
    • 8 years ago

    I checked AMD stock and its getting hammered by this news.
    a) The situation is worse then stated or b) AMD just became a good investment.

    I also found some Opteron 6200 benchmarks floating around,
    nothing exciting but should keep AMD in the game… power efficiency will tell us more.

    • ronch
    • 8 years ago

    I’m not sure it’s well-advised for AMD to be pointing fingers like that. After all, they were once one with GF and GF is still their main fab partner. They shouldn’t attempt to whine and burn bridges this way, which only results in tarnished relationships in the world of business. Hey, corporations aren’t robots… they’re actually run by human beings.

    Nonetheless, sources around the web seem to be in disagreement whether Llano’s design or GF’s 32nm process is to blame for poor Llano yields.

    [u<]Possible[/u<] arguments supporting GF's failure to get 32nm right: - The fastest mobile Llano chips are still hard to come by, suggesting yield issues. - Bulldozer may have been delayed due to yield problems and/or clocking problems caused by GF. Rev. B0 and B1 could well be hampered by GF 32nm. - Even desktop Llano parts aren't exactly everywhere (although this could simply depend on regions or other factors in the channel not related to GF). [u<]Possible[/u<] arguments supporting AMD's failure to get Llano right in terms of manufacturability: - Rumors suggesting early attempts at manufacturing Llano resulted in very bad yields, which still persist nowadays. Of course this could also be used to put the blame on GF. - Rumors suggesting AMD will halt Phenom II/Athlon II production this year. This is reckless on AMD's part if BD is experiencing yield problems at 32nm just like Llano. This suggests that BD's yield is ok on GF's 32nm and Llano's yield problems are all its own. I imagine these finger-pointing scenarios between AMD and GF aren't new. Just because they're two separate companies now doesn't mean they recently start acting this way. I'm sure they've been doing this internally amongst themselves from the very beginning. After all, the folks designing the chips aren't the same folks working in the fab.

      • WaltC
      • 8 years ago

      [quote<]I'm not sure it's well-advised for AMD to be pointing fingers like that. After all, they were once one with GF and GF is still their main fab partner. They shouldn't attempt to whine and burn bridges this way, which only results in tarnished relationships in the world of business. Hey, corporations aren't robots... they're actually run by human beings.[/quote<] AMD is required by law to specify why its previous guidance was wrong for the quarter (all traded companies operate under the same rules), else AMD could be sued into oblivion because of providing faulty and uncorrected earlier guidance. In talking about GF yields AMD was simply telling the truth as to why its revenues and margins were off the earlier guidance: IE, they had more demand for their products than GF yields allowed them to meet in the quarter. Of course, this was entirely the correct thing for AMD to have stated. You may not realize it, but all chip design companies experience yield problems from time to time, including Intel. All of them report it, too, as a rule, to explain why sales and profits failed to hit projections, if yield problems were the cause. Nobody is "burning bridges" of any type...;) This is common, actually. It's exactly what nVidia did during the nV30 fiasco--they blamed TSMC for their problems, IIRC--and even went so far as to hire IBM for a time to FAB some of their gpus. All of that is in the past for nVidia, of course, and nVidia and TSMC are thick as thieves today. It's all a part of doing business in the chip field. This will not affect GF/AMD relations at all--especially since GF knows what its yield issues are as well as AMD. I recall years ago arguing in some cpu forum somewhere with some guy who said that he had it on great authority that the AMD64 architecture was "flawed" and that it would "never hit 2GHz" because the "A64 architecture doesn't have the headroom." Oh, boy, did we argue--for what had to be weeks, on and off, about that...;) My position was that it was a much simpler problem of yield versus his contention that the problems were insurmountable because of insufficient "architecture headroom." I thought he was nuts--and he was--so you might want to think about that the next time some screwball wants to tell you that it's a matter of "architecture headroom" instead of yields. AMD integrated metal layers into its processes and whizzed right by the 2Ghz barrier as if it wasn't there. What happens when an architecture itself is so impractical that it cannot be made economically at various processes is what happened to nVidia's nV30 and Intel's Pentium 4: they get EOled and cancelled and replaced by a new architecture. Far, far more common than that, however, are temporary manufacturing hitches caused by entirely surmountable yield problems. This certainly looks like one of the latter situations to me. I'd put more stock into what AMD says at the moment than I would into Internet chatter and speculation.

        • flip-mode
        • 8 years ago

        TLDR

        • sschaem
        • 8 years ago

        Whats doesn’t add up is that llano has been in production since at least April 2011 with first run in mid 2010.
        And on July 15th 2011 AMD praised their Q2 quarter on llano success, no sign of yield issue and they guided higher for Q3 on llano.

        So what happen… did yield go down in august ?
        Most likely AMD didn’t tell that yield where terrible and was having huge problems.
        They where just hopeful for better yield, but it never happened. But that blind hope was translated to actual guidance.
        Something isn’t right in the way AMD disclose information as its very selective and deceptive.
        And they are going to pay the price for this strategic mistake for a long time to come.

        And what about the he risk with goflo: any chances they stop subsidizing AMD for its low yield design, and instead sell their capacity to a better client? I’m sure GoFlo wants to make money.
        And if AMD must pay per wafer, what happen to AMD margin? pay 100$ to goflo per CPU?

        note: The worse part in the announcement, no indication that the issue was address.
        So this is still a problem as of September 28 2011.
        It seem something need to change and change fast, AMD got no more cards in its deck.

      • flip-mode
      • 8 years ago

      It could be both. If all else fails, trust Charlie D.

        • NeelyCam
        • 8 years ago

        Charlie D is public enemy #2.

        Who is #1!!??
        .
        .
        .

        Chuck D.

          • CuttinHobo
          • 8 years ago

          To me, “Public Enemy #1” will always be one of the tougher opponents in the Road Rash games.

    • EV42TMAN
    • 8 years ago

    all i got from this is that AMD doesn’t know how to make large quantities for their CPU’s which is fine with me since i’m currently an Intel fan since AMD hasn’t made anything remarkable with the 939/940 socket ( and yes i’m including server cpus too)

      • clone
      • 8 years ago

      I’ll be switching to BullDozer if it performs, been an Intel/AMD buyer all along but nowadays nothing is exciting and all of it’s faster than need so waiting isn’t a problem and saves me coin.

      so long as it’s priced ok and performs I’m going AMD this time and my current system is will be fine for another year or two.

      found the biggest improvement I made in the last 2 years was going video and SSD hard drive anyway, CPU’s are boring comparatively speaking.

        • NeelyCam
        • 8 years ago

        [quote<]...SSD hard drive..[/quote<] Solid state drive hard drive..? Do you mean a hybrid?

          • clone
          • 8 years ago

          meh, my mistake should have said drive not hard drive.

          you knew this of course but …. meh.

            • NeelyCam
            • 8 years ago

            I was in a pissy mood – sorry about that.. Nonetheless, even SSDD is a bit redundant

    • chuckula
    • 8 years ago

    To punctuate the issue with Llano yields: Newegg is out of stock of the A8 3850. (http://www.newegg.com/Product/Product.aspx?Item=N82E16819103942)

    If Newegg doesn’t have it, a bunch of other retailers probably don’t have it either.

      • JustAnEngineer
      • 8 years ago

      I didn’t have any issues when I bought one a few weeks ago.

      • JustAnEngineer
      • 8 years ago

      Did you try Directron?
      [url<]http://www.directron.com/ad3850wngxbox.html?newsletter=09272011[/url<]

    • sschaem
    • 8 years ago

    AMD doesn’t deserve to be a public company if it cant even forecast 8 week in advance its operations.
    8weeks ago AMD publicly disclosed numbers that painted a whole different pictures.
    No issue what so ever with llano production, no issue with bulldozer or delays…

    They also clearly lied about the CEO search.

    AMD board hopefully own allot of shares of AMD, because this type of cover-up is not well accepted and AMD is going to get slammed.

    I also wonder if any insider trading happened. AMD lost a third of its value in the past 2 weeks, while intel gained a few %

    The issue might be much bigger then disclosed.

    In retrospect, llano first run was done in late 2010… almost a year now and they haven’t fixed the issue.

    bulldozer seem to be faced with the same problems, so even if bulldozer was a good design (it doesn’t seem to be beside very specialized server code) AMD wont be able to build them in any volume

    And with Radeon push to 2012 ….

    • Ryhadar
    • 8 years ago

    I think it’s kinda funny to read some of the comments for AMD to drop GloFo when GloFo [i<]was[/i<] AMD until a few years ago.

    • MadManOriginal
    • 8 years ago

    Didn’t NeelyCam post in these comments? (And get downrated by the AMD fanboy brigade for being right.) Am I just dreaming? I hope so, because I will lose what little faith I have in the internets if someone other than he deleted the post and thread.

      • khands
      • 8 years ago

      The issue is that you have faith in the internet.

      • jensend
      • 8 years ago

      You might try looking around before posting paranoid “OMG TR EDITORS ARE A PRO-AMD CABAL WHO DELETE ANTI-AMD COMMENTS” conspiracy theory accusations. NeelyCam posted in the shortbread, not here. (He was downrated because he had nothing worth saying and- in his own words- just wanted to have the “first piss,” not “for being right.”)

        • NeelyCam
        • 8 years ago

        First piss was a bonus. The main reason for posting was to point out that I was right and AMD Thumbdown Brigade was wrong.

        Sorry, but these kind of posts are bound to happen when people downvote others just because they have dissenting points of views. Those getting downthumbed unjustifiably will get pissed off and come back to flaunt it if it turns out they were right.

        My post was a result of AMD Thumbdown Brigade’s hypocritical bias. Blame them.

          • Silus
          • 8 years ago

          LOL AMD Thumbdown Brigade! That suits them fine since that’s basically the only thing they do πŸ™‚

            • NeelyCam
            • 8 years ago

            If a post of mine is above -2 it’s probably worth reading, since there is a basic handicap of about -3 on all my posts.

      • NeelyCam
      • 8 years ago

      Don’t worry; my post is still happily racking up thumbdowns in today’s Shortbread.

        • dpaus
        • 8 years ago

        I regret that I have but one vote to give…

          • NeelyCam
          • 8 years ago

          Hmm.. maybe TR should start selling extra thumbs? Say, for $5 you can buy 100 thumbs, and you can use a total of 10 per comment.

          That could bring in more cash than ads.

            • JustAnEngineer
            • 8 years ago

            +1.

    • chuckula
    • 8 years ago

    Bulldozer’s large die size IS having a negative impact on the situation, even though yields of BD itself are fine. That’s because GloFo has limited production capacity and there are opportunity costs associated with production.

    Let’s assume that BD has 100% yield. If BD could have been designed with approximately the same die size as Llano (~250 mm^2) instead of 315 mm^2, you’d have about 28% more chips per wafer in BD. That’s about 28% more wafers that could be devoted to Llano production to satisfy the OEMS while maintaining the same volume of BD production, or some intermediate balancing thereof.

    If all GloFo had to do was produce Bulldozer, then the problem wouldn’t be that bad, but factor in the other demands on production and BD causes some problems. It’s by no means the worst problem that GloFo faces, but it isn’t helping things either. Conversly, you could say that Llano is hurting Bulldozer since lots of wafers need to be put into the queue for Llano production instead of BD production.

      • shank15217
      • 8 years ago

      BD is larger because it has 16MB of cache because its primarily a server chip. BD is also going to sell for a much higher ASP, so blaming AMD for designing a ‘big’ chip is baseless. GLOFO has to get their shit together and improve 32nm yields because that’s what fabs do. AMD isn’t GLOFO so AMD has all right to blame it’s fab partner for delays. AMD should look into other FABs for future products.

        • Silus
        • 8 years ago

        Yet some “other” company, with a big chip that will be used through several markets where that “other” company works in, is criticized for creating a big chip and if said “other” company criticizes the fab for poor yields, it’s their own problem and the fab is innocent! Double standards at their finest! πŸ™‚

          • theonespork
          • 8 years ago

          Ummm, you do realize that the “other” company you are discussing is Intel, and Intel owns all their own fabs. Inherently, criticizing one is criticizing the other in Intel’s case.

            • Silus
            • 8 years ago

            There you go again with your assumptions…I wasn’t talking about Intel.

            • theonespork
            • 8 years ago

            Really? I am curious as to who you are speaking of? NVIDIA does not seem to fit the bill, as we are discussing CPU’s. IBM? Oracle? Fujitsu? ARM?

            • NeelyCam
            • 8 years ago

            He was talking about chips in general – not just CPUs – and his point is very valid.

          • HighTech4US2
          • 8 years ago

          Pay the man Shirley

      • tviceman
      • 8 years ago

      Nvidia probably has the largest die sizes in the chip industry, they share TSMC with dozens of other chip companies, and yet they don’t seem to have supply issues. Both GF100 and GF104 are bigger than Bulldozer.

      In other words, Bulldozer’s size is no where close to having a noticeable impact on their current situation.

        • bcronce
        • 8 years ago

        The cookie cutter design of a ton of shader units allows minor defects in GPU cores to allow for a still workable chip.

        Large cache of the BD means a defect in the cache/core and either you’d have to disable much larger portions of the cpu or toss the chip

        • OneArmedScissor
        • 8 years ago

        Nvidia has huge problems with their huge chips, over and over. I have no clue what you’re talking about. Their largest chips are comparitively [i<]extremely[/i<], may I repeat, [i<]extremely low volume.[/i<] AMD is making dual-die server CPUs for everything from low power to super computer multi-socket servers. If you think it's going to take less silicon to fill the demand in that market than the tiny niche that Nvidia's highest end serves, then I don't know what to tell you.

    • Silus
    • 8 years ago

    Yeah right! The typical whining! A chip is designed to fit a certain fab process. Who designs the chip for that process has as much responsibility as the fab that uses it and maintains it, when problems arise.

      • theonespork
      • 8 years ago

      This is an inaccurate statement. A properly designed chip can and sometimes will have yield issues due to immature fab processes and/or untweaked fab processes. These problems can be compounded when the silicon is pushed out on bulk silicon and not a truly custom die, which I believe Llano qualifies as.

      We are all waiting patiently for Bulldozer, some to praise its glory and root for the underdog, some to bury it as a clear failure from an incompetent competitor, and some simply to blister and lampoon either side as their inconsistent arguments and inane fanboy jingoisms warrant.

        • Silus
        • 8 years ago

        And your assumption is, that this is a properly designed chip (which is where your argument holds any water) ? Based on what exactly ?

          • theonespork
          • 8 years ago

          I neither made nor implied an assumption that the chip was properly designed. Your statement was a general statement about a particular product. The general statement is inaccurate. Llano may well be improperly designed, but you provide no evidence of this.

            • Silus
            • 8 years ago

            And I just pointed out that your labeling of my argument as “inaccurate” is inaccurate too, because it only works IF the chip is properly designed, which is also something you have no evidence for.

            • cegras
            • 8 years ago

            And you do not have evidence to the contrary.

            • flip-mode
            • 8 years ago

            While Silus’s posture is certainly annoying, the rumor that BD yields are fine and Llano yields are not and that AMD is looking to transition away from Llano to Trinity so quickly does, if not prove, strongly indicate that there’s something not right about Llano.

            Hearing it from an Nvidia fanboy can make even good logic seem tainted.

            • NeelyCam
            • 8 years ago

            This. Silus is right. Getting good yield is the responsibility of both the fab and the design team.

            Remember AMD’s 4770? ATi folks knew how to do this well on TSMC – they used 4770 as a test chip to get the yield issues figured out (while NVidia was running around like a headless chicken). As a result, AMD completely controlled the graphics market with 5850 and 5870 for six months.

            This time around, the graphics team was working on a whole new process… first time with SOI. My guess is that they didn’t have the experience to make a good-yielding design on SOI, and GloFo process ramp issues made things worse. They’ve probably learned a ton by now, and Trinity will yield better. Meanwhile, Llano will become a “test chip” just like 4770 did.

            All this is fine when your competition is on the same process and learning curve (i.e., in discrete graphics). But when your competition is two years ahead in the learning curve and uses a very mature process, you’re toast.

            • Silus
            • 8 years ago

            You find “logic” and “well known facts in the industry” as “annoying”, but only because I said them. How sad can you get ?

            • flip-mode
            • 8 years ago

            Calm down a minute. Notice it wasn’t the facts and logic that I was labeling annoying.

            • jensend
            • 8 years ago

            Anything Charlie says “strongly indicates” nothing– except the fact that Charlie likes to talk. Show us a source for these “rumors” of BD/Trinity yields being much higher than Llano that has some credibility and then maybe your claims that these rumors bear more credibility than AMD’s official statements to shareholders will be a little less laughable.

            Charlie can make random stuff up any day and there are no consequences when he’s wrong. If AMD lies to its shareholders in official statements like this, they’re legally liable and open themselves up to all kinds of lawsuits. I think their statement that the problems are at GloFo bears a heck of a lot more weight than Charlie’s latest random BS.

            • NeelyCam
            • 8 years ago

            You’se an nvidia fan?

            Charlie’s accuracy on early rumors is pretty good. Not perfectly accurate, but semiaccurate… Just because he hates NVidia and you don’t doesn’t mean his information isn’t valid.

            And if he outs his sources, there wouldn’t be any rumors anymore. Tech rumor business relies on keeping sources confidential.

            [quote<]Charlie can make random stuff up any day and there are no consequences when he's wrong.[/quote<] This is true, but if he makes up stuff, he soon loses his credibility. So far, I'd say he's been right more often than wrong, and a lot of times when he's wrong, it's because company plans have changed.

            • theonespork
            • 8 years ago

            Your logic is flawed. You made a blanket statement which implies a rule that can be followed. I stated this too be “inaccurate” which has thus far proven a true statement. Had you stated that in this case it was clear AMD was as much at fault as GF I would have asked for proof, but I would not have stated your argument was “inaccurate”. Either you were baiting in your original comment or you are playing CYA now.

            • NeelyCam
            • 8 years ago

            Here’s his original comment:

            [quote<]A chip is designed to fit a certain fab process. Who designs the chip for that process has as much responsibility as the fab that uses it and maintains it, when problems arise.[/quote<] There are two statements here that could be considered "rules". First: "A chip is designed to fit a certain fab process." This "rule" in most cases is true. Not in all cases, but in most cases we're concerned about (CPUs, GPUs..) If you want to include non-CPU, non-GPU chips in this argument, then yes - you could say it's inaccurate. Second: "Who designs the chip for that process has as much responsibility as the fab that uses it and maintains it, when problems arise." The only thing potentially inaccurate here is "as much" - implying that the responsibility is shared equally. This can be argued either way. Note that he never said he was talking about Llano (or BD) in particular - it was you who thought that's what he was talking about, and started attacking him about it. Implying he was talking about a particular chip is the most inaccurate aspect of the whole debate here.

        • sschaem
        • 8 years ago

        Fact

        – llano first run was in late 2010. (Demo done in October 2010)
        – 8 weeks ago AMD publicly stated that llano had no issue in its production and gave only positive forward prediction

        AMD / goFlo have not been able to control 32nm manufacturing for over a year and never said anything publicly.
        AMD covered up this fact, and no its not going to be resolved soon.

        I smell a class action lawsuit if some big boys where heavely invested in AMD, because this is pure deception on their part.

        The same issue will plague bulldozer… we already seen this with the delays. And again, 8 weeks ago AMD stated clearly their Interlagos schedule… yet yesterday they said interlagos shipped latter then expected.

        AMD is at fault for not using its Fab capabilities and designing chip that cant be manufactured.

    • Arclight
    • 8 years ago

    GloFo is AMD’s spin off, hence they need to turn it into a sucess story. I’m sorry they got a bad start, hopefully they can turn this around. What everybody is forgeting is that AMD actually made money, ha!

      • chuckula
      • 8 years ago

      AMD may technically have made money, but this is sort of a shell game they are playing financially. Back in 2009 when AMD was losing money by the bucketful the ATI side of things was doing much better than the CPU side… you could have said that ATI actually made money then!

      Until AMD gets full first-class support for all of its main revenue producing chips from either more than one foundry or until and if GloFo gets its act together, AMD is sort of a hostage. Now, AMD has nobody to blame but itself since the AMD board decided to do the spinoff in the hopes of making things more profitable.. you have to take the good with the bad in your decisions.

      Getting your chips from multiple foundries sounds nice, but it is difficult to do. You can’t just take a single chip design and throw it at different foundries even at the same node and expect them to start pumping out functioning chips quickly. When it comes to CPU and integrated GPU parts, AMD looks to be stuck with GloFo for quite some time to come.

      The pure GPU side of things is looking a little better if only because AMD’s main competitor Nvidia has to contend with the exact same delays from foundries that AMD does.

        • Action_Parsnip
        • 8 years ago

        The thing is the backers of Global Foundries have more money than god. Factor in that Glo.Fo. is expanding very aggresively and you have a company that will be going places, but there will be blood along the way.

          • NeelyCam
          • 8 years ago

          That blood will be AMD’s. GloFo doesn’t have to care – Abu Dhabi oil money will last a long time. But, AMD doesn’t have that lifeline.

        • just brew it!
        • 8 years ago

        I was under the impression that GloFo is one of the few foundries that has experience with SOI. AMD may have boxed themselves into a corner by betting the farm on an alternative fabrication technique, as this probably limits the foundries they can partner with.

        SOI wafers are also a lot more expensive than bulk silicon, which means the low yields are even more painful.

        Anyone know if construction of the new fab in NY is still on schedule? I found [url=http://www.fab2construction.com/<]a link[/url<] which seems to indicate that it is supposed to be up and running "early 2012", but the site does not appear to have been updated since April.

      • Peldor
      • 8 years ago

      Unless there are other details out, this is just revenue and margin. Profit is another matter. I expect they probably did stay in the black though.

        • HighTech4US2
        • 8 years ago

        When companies warn on revenue misses and lower margins they usually follow with a profit loss. Expect this quarter to be in the red.

          • NeelyCam
          • 8 years ago

          I’m with HighTech. Loss likely this quarter. Anything remotely high-margin has been hampered by GloFo. Brazos, while being in high demand, doesn’t have such high margins… and fixed costs (all those engineers..) have to be paid regardless of how well the chips sell.

      • HighTech4US2
      • 8 years ago

      And AMD will lose money this quarter and for as many quarters in the future until they fix the 32nm problem.

    • chuckula
    • 8 years ago

    SemiAccurate can have issues, but they have called this one on the accurate side as opposed to the whacko side, here’s a quote:
    [quote<]To make things more interesting, SemiAccurate moles tell us that the problems are related to Llano in particular, not the 32nm process. Bulldozer and Trinity, the next two 32nm parts coming out of GloFo, are yielding much better, far above Llano levels. Why does one design have massive issues while another not? Some people have said that the GPU is to blame, but it isn’t, Trinity doesn’t suffer from the same ills. Short story, no one is talking yet.[/quote<] Apparently yields of fully operational A8 chips are in the 5 - 15% range(!) This is probably why AMD is pushing out Trinity as fast as humanly possible. It would likely let Llano linger on another quarter or two if Llano yields were fine and the OEMs weren't complaining about supply. Source: [url<]http://semiaccurate.com/2011/09/28/amd-misses-bergman-scores-all-in-one-day/[/url<]

      • flip-mode
      • 8 years ago

      Chalk one up for Charlie. Too bad he’s so wordy. Word count in his articles is usually at least twice what it needs to be. If he were a poet then I wouldn’t complain, but he is no poet.

        • NeelyCam
        • 8 years ago

        Yeah – with his grammar, a Pulitzer is out of reach..

      • NeelyCam
      • 8 years ago

      Interesting.. so, whoever process-shrunk the core took some short cuts. Or, maybe it should’ve been re-designed from scratch to yield well, but schedule and resources wouldn’t allow it – had to get stuff out the door asap to battle SB.

      An alternative would’ve been nixing Llano altogether, and focus on pulling in Trinity.

      • HighTech4US2
      • 8 years ago

      You mean Flip-Flop charlie is accurate. His site is named SEMI-Accurate and that is usually what his posts are. Some accurate some in-accurate.

      Here is a prime example:

      [url<]http://semiaccurate.com/2011/09/28/amd-misses-bergman-scores-all-in-one-day[/url<] ------------------ Quote(s): To make things more interesting, SemiAccurate moles tell us that the problems are related to Llano in particular, not the 32nm process. Bulldozer and Trinity, the next two 32nm parts coming out of GloFo, are yielding much better, far above Llano levels. Followed by an update today: More sources have come forward and said that Trinity is indeed having yield problems, but not as bad as Llano. Trinity is said to have started off better than Llano, and still have 2Q or so of optimizations to go before launch. That said, it is still more problematic than Bulldozer, but it is not the fault of the GPU. ------------------ Usually his in-accurate ones never get an update even when proven to be false so for him to actually update one it must be really bad for AMD.

    • Tristan
    • 8 years ago

    It seems, full 8 core Bulldozer will be very rare

      • jjj
      • 8 years ago

      BD is supposed to have ok yields.

        • wingless
        • 8 years ago

        Source? I would love this to be true.

        • sschaem
        • 8 years ago

        Untrue, AMD just stated that interlagos was delayed because they cant ramp up production.

          • NeelyCam
          • 8 years ago

          They did say that Interlagos was delayed, but they didn’t state the reason. Concluding that it’s GloFo yield issues is factually inaccurate.

            • sschaem
            • 8 years ago

            This is the only logical conclusion using all the information available.

            This disclosure is a revision from the Q3 forecast delivered July 15h.
            AMD expected Interlagos to ship in volume mid august, instead it ship in limited volume September7.
            3 weeks later, still no volume shipment. GoFlo cant ramp production.

            Whats your view on this ? GoFlo has been producing bulldozer chip since early September at full volume and AMD is sitting on top of tens of thousands of Opteron processors but dont want to sell them… ?

            • NeelyCam
            • 8 years ago

            I agree with you that yield issue is a very possible explanation. I was merely pointing out that AMD didn’t say that, and your comment made it sound like they did.

            EDIT: .. I was [i<]trying[/i<] to point that out, but looking back at my comment, it didn't say it too well...

            • sschaem
            • 8 years ago

            I cant deny, AMD didn’t state anything directly about BD yield…

            I often write extreme statement just looking for someone to put me in my place with a good rhetoric.
            Sadly with AMD this is not happening πŸ™ a bunch of thumb down only encourage me πŸ™‚

            In this case, I’m still thinking that If yield for 32nm BD was ok. they would have stated so..
            Because that announcement sounded like a suicide note, and the timing was oh so ‘perfect’ too.

            • NeelyCam
            • 8 years ago

            [quote<]Sadly with AMD this is not happening πŸ™ a bunch of thumb down only encourage me :)[/quote<] That's about 50% of why I keep coming back. πŸ™‚

        • HighTech4US2
        • 8 years ago

        Take that “ok yields” with a grain of salt as the only source is from the AMD-biased Char-Lie of Semi-Accurate.

        • sschaem
        • 8 years ago

        Bulldozer as been in production since what, late august? where are the chips after one full month of production…

          • flip-mode
          • 8 years ago

          “In production” does not at all mean “available for purchase”. There are many stages to production and some of those stages happen not just in different buildings but different countries. Furthermore, new products don’t go straight from the production line to the retail shelf. The manufacturer usually stipulates when retails sales can commence. This is so an adequate stockpile of product can be available so that products don’t go “out of stock” from one day to the next. Big retailers will even buy huge orders – months worth of supply – at a time.

          So, in other words, we cannot deduce anything about Bulldozer yield from retail availability until the product is actually launched. Then after a month or two of retail data we will be able to see what availability for Bulldozer looks like. If it is going out of stock frequently then AMD (GloFo) is probably having trouble producing enough.

            • Coran Fixx
            • 8 years ago

            My personal theory is that they have plenty of Bulldozers, but when they put them in motherboards and power them up they sound like when the hyperdrive fails on the Millennium Falcon.

            It is an improvement over last month when they sounded like R2D2 taking a direct hit.

            • flip-mode
            • 8 years ago

            That would be awesome. If Windows played that sound at BSOD I’d be constantly trying to crash my computer.

          • Action_Parsnip
          • 8 years ago

          Doesn’t it take like 6-8 weeks to go from input materials to finished wafer? Plus there’s packaging and testing.

            • intangir
            • 8 years ago

            “Production, start to finish, on a wafer, is more like 12-13 weeks.”

            [url<]http://blogs.amd.com/work/2011/09/07/the-start-of-a-new-era/comment-page-1/#comments[/url<]

      • OneArmedScissor
      • 8 years ago

      That’s a weird conclusion to jump to. Low yields doesn’t mean less of X part of a chip if Y part of it is the problem. It doesn’t even have to mean more chips with disabled parts. Some may just come out not functional at all, while the rest are fine. See: TSMC 40nm debacle. Sometimes there is just something wrong on the manufacturing side that is hard to pin down.

      But that has nothing to do with this. The problem has been Llano’s overly ambitious GPU for an entire year, which was exacerbated by only making the quad-core version with the extra huge GPU for the longest time.

        • sschaem
        • 8 years ago

        No interlagos also suffered yield issues.
        Read AMD press release. 8 weeks ago interlagos was believed to be on track, but AMD couldn’t ramp up production and thats why it shipped late.

        AMD/goFlo has been struggling with 32nm yield since mid 2010. Remember AMD publicly showed llano in october 2010.

        AMD earlier this year was in a panic and had to redo its contract with goflo about who pays for dead chips…

          • NeelyCam
          • 8 years ago

          Nowhere has it been said that the delay is somehow attributed to GloFo yields. BD delay could be a design issue.

        • willmore
        • 8 years ago

        Correct me if I’m wrong, but doesn’t Llano contain the first AMD/ATI GPU ever made on an SOI process? Plus, it was a new node to boot. Hold cow, it’s amazing any parts worked.

          • NeelyCam
          • 8 years ago

          Yes, but Charlie says it’s not the GPU – it’s the CPU.

    • chuckula
    • 8 years ago

    [quote<]Additionally, 45nm supply was less than expected due to complexities related to the use of common tools across both technology nodes. [/quote<] Looks like they jumped on 32nm and forgot that you still need to sell 45nm parts when your 32nm parts are not enough to fill demand...

      • sschaem
      • 8 years ago

      Speculation: AMD was forced to switch production to interlagos to fulfill Cray contract.
      and they burned capacity with low yeld bulldozer…

      If AMD was able to delay interlagos by two month, things would have been A.OK

      Cray might actually have been a big influence on the bulldozer design. Great for Cray, bad for the PC market.

    • wingless
    • 8 years ago

    Well it is indeed Global Foundries’s fault. Do they actually understand that the fate of Intel’s ONLY real competitor lies squarely in their hands? AMD should find new foundry partners if they haven’t already done so.

      • shank15217
      • 8 years ago

      Yup, AMD needs to diversify their partners and maybe look into keeping their GPU line away from GLOFO for the time being.

      • bcronce
      • 8 years ago

      It’s not easy to find a new fab. A fab costs about $6bil to create and that doesn’t include R&D and operational costs.

      Because of this, there are not many choices. Global Foundries and TSMC are about your only options. Everyone else has their own in-house, like Samsung and Intel.

      The biggest issue is everyone sees Intel at 32nm and moving to 22nm soon. Intel kept their R&D during this economy, Global/TSMC/etc all made large cuts to R&D.

      Intel may have evil lawyers/marketing/sales, but their R&D is top notch.

      • khands
      • 8 years ago

      They’re working on it, most of their parts at 28nm should be manufactured both at TSMC and GloFo

        • NeelyCam
        • 8 years ago

        TSMC 28nm is gate-last… is GloFo 28nm gate-last?

Pin It on Pinterest

Share This