Sign in with
Sign up | Sign in

Sandy Bridge Bug 2X Costly as Pentium Math Bug

By - Source: Tom's Hardware US | B 75 comments

Intel's surprising note that it has discovered a "circuit design issue" in the 6-series chipset is reminiscent of the 1993 FDIV bug. Intel's quality control may not be as fail-safe as the company claims.

The FDIV bug was a catastrophic design flaw in the original Pentium (P5) processor that caused certain floating point divisions to produce false results. It was a rather controversial problem and extremely rare as only one in 9 billion operations was believed to be affected by the flaw, but Intel was forced to recall and correct the issue. The company initially expected the cost to be in the neighborhood of $300 million, but ended up paying about $475 million in total.    

We don't know much about the Sandy Bridge SATA bug announced today, other than the chipset could see its SATA connection deteriorate and cause the performance of SATA devices such as hard drives decline. Intel says not many consumers are affected and it has already begun manufacturing new Sandy Bridge devices. However, the company expects the issue to cost about $700 million, which isn't exactly a number for a minor issue. Semiconductor companies are traditionally conservative with their estimates how much such issues could cost in the end and there is a good chance that the SB problem will cost Intel nearly twice as much as the 1993 FDIV bug.

We should note that Intel has an elaborate structure of quality control in place: there are more than 3000 engineers worldwide who are testing new chip designs usually for at least 9 months before a chip is commercially released. This structure has been created following the FDIV bug specifically to avoid scenarios such as the announced Sandy Bridge circuit design "issue." With a $700 million bill and the stock market fallout still to be seen, Intel may be looking into ways on how to improve its quality control system -- especially since the company is trying to figure how to make its way into other high-volume markets such as mobile phones and consumer electronics.

There have been some reports from the side of AMD ridiculing Intel's mistake, but we should be fair and recognize that microprocessors are highly complex marvels and engineering and AMD has made mistakes as well - such as the TLB bug in the company's first quad-core server processor - the 2007 Opteron CPU with Barcelona core.             

Display all 75 comments.
This thread is closed for comments
  • 1 Hide
    bv90andy , January 31, 2011 6:44 PM
    So much for their record profits. I know where those will be spent. Too bad. because this will finally be payed by the consumer.
  • 1 Hide
    eddieroolz , January 31, 2011 6:45 PM
    That's some serious dough there. My guess is that it will run into the billions towards the end.
  • 0 Hide
    yao , January 31, 2011 6:59 PM
    when can I buy my pc?
    I was thinking about buying one, now this news come out.. no way I will buy in the near future.
  • 1 Hide
    silverblue , January 31, 2011 6:59 PM
    Sorry, but we're talking 1993 money as compared to 2011. It's way worse than 2X.
  • -2 Hide
    geekapproved , January 31, 2011 7:00 PM
    Definitely billions. And what about the OEM's and mobo makers, they going to have to absorb all the RMA and shipping costs for replacement parts and replacement labor. What about the Mom and Pop shops, they are seeing record sales right now because of SB release, now they can't sell any more till April and nobody but the super budget minded are really buy AMD right now, so, there goes those record sales.
  • -2 Hide
    geekapproved , January 31, 2011 7:01 PM
    Wow, you think after the P55 cpu socket issues they would have bumped up the QC a little. Guess not, they are too worried about aquiring Mcafee. lol
  • 6 Hide
    zelog , January 31, 2011 7:07 PM
    silverblueSorry, but we're talking 1993 money as compared to 2011. It's way worse than 2X.

    You mean "it's way better than 2X"
  • 0 Hide
    silverblue , January 31, 2011 7:08 PM
    Depends on your interpretation. :)  How much is $475 million 18 years on?
  • 0 Hide
    Onus , January 31, 2011 7:16 PM
    The mainstream market, where the vast bulk of systems are sold, won't even notice. That market is still buying S775 systems, e.g. "Corporate Stable" models. This will inconvenience, piss off, and maybe alienate some enthusiasts, and maybe the top percent or two of the high-end pro market, but nothing more.
  • 0 Hide
    accolite , January 31, 2011 7:20 PM
    Looks like it's intel's turn to go the amd route.
  • 6 Hide
    JamesSneed , January 31, 2011 7:20 PM
    Take your time Intel. AMD don't fubar Bulldozer. This may be an interesting year.
  • 1 Hide
    silverblue , January 31, 2011 7:21 PM
    Probably true. Depends on how much of a fuss the trade computer mags kick up.

    I got myself tied up with the "it's way worse than 2X" earlier; if I'm correct, $475 million would be worth about $700 million nowadays, so we're still some way short of it being twice the pain for Intel, even at $1 billion. It's still lower than the AMD and nVidia settlements and they won't have any real issues paying it.
  • -2 Hide
    zkevwlu , January 31, 2011 7:25 PM
    Assuming a prevailing annual interest rate of 3% (which is really really damn conservative), a 475 million dollar investment 18 years ago is worth 809 million dollars today. So no, there is no way in hell this debacle is going to cost twice as much as the FDIV bug.
  • -8 Hide
    milktea , January 31, 2011 7:27 PM
    Intel (or at least a few at Intel) did this on purpose to give AMD some breathing room. Things a more exciting with competition. :p 
  • -2 Hide
    nforce4max , January 31, 2011 7:28 PM
    I got a few pentium 1 cpus in my collection including a few engineering samples :)  some of which are not even 100mhz lol
  • 2 Hide
    wiyosaya , January 31, 2011 7:29 PM
    Is this what happens when the bottom line / shareholders are the only thing that counts and products are rushed to market without adequate testing?
  • 2 Hide
    voodoobunny , January 31, 2011 7:31 PM
    "Reminiscent of the FDIV bug"? As in, denying that a problem existed for months until the recall was far more visible and the product's reputation was in tatters? I think not; Intel is shipping *far* more units now than they were in prehistoric times (sorry, I mean 1993) and they will probably actually get some positive press over this for being so quick to act.
  • 7 Hide
    tpi2007 , January 31, 2011 7:32 PM
    This is worse than the FDIV problem of the original Pentium for a simple reason: Intel deliberately prevented another manufacturer from making chipsets for their CPU's.

    If Nvidia was still around for example, people could still buy a Nvidia motherboard to go with the Intel Sandy Bridge CPU. As it is, Intel loses twice: people will hold off from buying not only the affected motherboards (which accounts to ALL of them right now), but it will also not sell the brand new CPU's.

    I think it's Nvidia right now that must be laughing and not exactly AMD.

    Intel wanted to be alone in the chipset business, now it has to take on all the inherent risks of that decision too. There was a very good reason IBM told Intel in the beginning that it would only order chips from them if there was a second CPU licensee - they wanted the supply risk at a minimum.

    EDIT: Newegg just took offline not only all LGA 1155 based motherboards, but also all Sandy Bridge Core i5's and i7's.
  • 2 Hide
    Anonymous , January 31, 2011 7:34 PM
    It will come out of their Petty Cash account.

  • -1 Hide
    Kileak , January 31, 2011 7:38 PM
    I'm glad you took the time to tell everyone reading the article that AMD has also made a mistake at some point in time because if not, obviously everyone would've thought that they're perfect...

    Great way to detract attention from the culprit.

Display more comments