Yield 2390

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

Digitimes has leaked some very interesting information about Hammer production (bolded print our emphasis):

Unit sales of Advanced Micro Devices (AMD) Athlon 64 (K8) processors are expected to grow 50% sequentially in the second half of this year, according to sources at Taiwan motherboard and chipset makers.

Improved yield rates for K8 processors and AMD’s recent price cuts, about 20% on average, for the Athlon 64 processor lineup, are two key factors driving second-half sales, stated the sources.

Unit sales of K8-based motherboards accounted for only about 5% of total motherboards sold in the first half because AMD was unable to improve the yield rate of the K8 processor, and the prices of the CPUs were relatively high, the sources said.

However, the sources noted that sales of K8 processors started gaining momentum in July and should continue to expand at a rapid pace in August, following the July 26 price cuts.

So now the truth comes out. AMD simply couldn’t make a lot of Hammers.

We have long suspected something like this has been the case.

The reason for that is very simple. Whenever AMD has this kind of problem with one of their products, they keep doing the same thing:

  • They stop talking much about the product.
  • When asked, they deny any problems with the product until they’ve fixed the problem.
  • Once the problem has been fixed, then they’ll admit to it.

    They did it with Thoroughbred As. They did it with the initial Opterons. We see it here.

    So fully expect AMD spokesmen to cheerfully admit former yield problems in the next days/weeks ahead.

    What About Intel?

    There sure aren’t too many 3.6GHz Prescotts floating around, are there? 🙂

    At the moment, only two Pricewatch vendors are making any pretense of being able to get you one, and only one says they can send one out in a couple days.

    Intel reacts to these situations a bit differently than AMD. When they have this kind of problem:

  • They officially deny the problem.
  • Almost immediately thereafter, the problem gets unofficially confirmed through other channels, or indirectly gets confirmed officially.
  • They then officially release a product that unofficially isn’t (or barely is) available until they can make it.
  • Once they can make it, they still don’t say anything about it officially.

    For instance, in the case of Prescott, when only low-speed Prescotts were initially available, Intel said “there’s no problem,” but at the same time introduced 3.4GHz Northwoods, plus indicated in a rather separate announcement that they were going to implement a new electrochemical brew in 2005

    Neither approach is praiseworthy, but both have advantages and disadvantages.

    With Intel, you usually don’t get a lot of advance warning that something’s wrong, but once that becomes clear, you can usually get a clue as to what the problem likely is.

    With AMD, you usually get more advance “warning” that something is wrong, but you’re left guessing as to what that something might be.

    Still Constrained

    The Digitimes report says that production of Hammers ought to be up 50% in the second half of the year over the first. That figure is rather consistent with earlier AMD estimates of production, and indicates production of about 1.5 million Hammers per quarter.

    That’s still not terribly good, and represents only about 20% of current AMD CPU unit production.

    We suspect that the bottleneck now will be 130nm Hammer capacity. AMD will probably not want to bear the expense of converting current production lines from XP to Hammers for just a couple quarters. All indications now are that AMD will try to seriously ramp Hammers end of Q1, early Q2 2005 from around 1.5 to about 4 million CPUs.

    These intentions could well change if AMD has problems with 90nm technology, which could be one or both of the following:

  • A Prescott-type problem (i.e., AMD can make medium-speed 90nm SOI chips OK, but not very fast ones); and/or
  • A 90nm yield problem (i.e., AMD cannot get high yields from 90nm SOI chips.

    If it’s the first (an indicator of that would be no 90nm 2.6GHz CPU next October), we’ll see an AMD version of Prescott’s release at lower speeds.

    If it’s the second (an indicator of that would be significant price premiums on any 90nm chips released), it’s more likely AMD will bite the bullet and convert more 130nm capacity over to Hammer.

    We find the first more likely to be the case.

    Forget About The Future, What About Now?…

    Forget About The Future, What About Now?

    Many reading this probably want to know first and foremost: Will this mean current Hammers are going to get a lot cheaper soon?

    Maybe, but you probably won’t have AMD to thank for that. AMD will probably want to keep its price parity policy with Intel intact, so price decisions will end up in Intel’s lap.

    On the one hand, the only price cuts Intel had scheduled for this year are in late August (AMD has already made the matching cuts Intel is expected to make), and another late in the year, ostensibly when the 4.0GHz Prescott is supposed to come out.

    Since Intel has announced the 4GHz Prescott won’t be out until next year; under normal circumstances, this would make a late-year price cut doubtful.

    However, Intel has also announced that their inventories have been building, which often leads them to initiate a price cut, new product or not.

    If I had to bet today, I would bet that Intel will make an additional cut, perhaps in late September or early October.

    How much waiting will help a potential Hammer buyer depends on what you plan to buy. If you want an Athlon 3000+ (whether socket 754 or 939, the socket 939 versions should be out by around then), any late-year price cut will only amount to about $15.

    On the other hand, if you’re looking for a higher-end Hammer (perhaps a 90nm socket 939 Athlon 3500+ will be available by November), you might save more like $100.

    It’s a little too early to make a call on this, not so much for price as to whether the products will be worth buying. Will 90nm Hammers be available? At what speed grades? Can they overclock much?


  • Discussion

    Leave a Reply