Rely On ATI?

Add Your Comments

Got this email recently:

Is there any reason that you know of that AMD keeps beating a dead horse with this K10? It is obviously a disaster and they just keep trying do something with it. I just can’t fathom why this company does not just stop what they are doing, live off of the ATI side for a year or so, and create a new design from scratch. They have wasted 4 years or more now in this fiasco and it is just doing nothing but killing them. I don’t want AMD to go away, and I will support them on the video card side since I like their products, but this cpu business is going to hell in a hand basket and they just keep rowing the boat down the same rough river. Is this something you might post on the website? Because I think I am not the only one scratching their head wondering what the hell is this company trying to do.

Well, I have been getting inquiries along these lines, so here goes:

Beating A Dead Horse Beating a dead horse is utterly insane, of course, if you have a live one. But what do you do when you don’t.

What you do is what AMD is doing, which is:

  • Get another horse, preferably one alive and breathing, if not quite in racing shape (45nm).
  • Put a new harness on the really old nag that was headed to the glue factory and put it through one last round (new lines of 65nm X2s).
  • Pretend for the benefit of the rest of the world that your dead horse is just sleeping and just needs a little time and encouragement to wake up (K10). You get a new saddle for it and talk a lot about how great it is (780G). By doing this, you hope that at least some of the betters in the stands don’t notice that the horse has passed its expiration date.

    In short, if you have lemons, make lemonade. Not like you have anything better to do, which brings us to the next point:

    Why Not Rely On ATI Until You Get A New Horse? To keep this short, here’s three big reasons why this doesn’t work:

    1) Tale wagging dog: AMD has three significant revenue producing divisions: Computing Solution, Graphics, and Consumer Electronics. For our purposes, it’s close enough to say that the first is the “old” AMD, and the other two are the “old” ATI.

    How much money do the two bring in? Let’s look at their revenues from last quarter:

    “Old AMD”: 1,402 million dollars
    “Old ATI”:      368 million dollars

    Yes, the ATI figures are a bit sickly, but it’s not like they’ll double or more anytime soon.

    AMD currently gets 80% of its money from making CPUs, good, bad or indifferent. AMD shutting down its CPU business and relying on the “old ATI” would be like Intel shutting down its CPU operations and relying on just chipset sales. Besides, once you strip out all the accounting games, the former ATI isn’t doing well, either, it actually lost more real money than the CPU people did last quarter.

    For better or for worse, as goes the CPU division, so goes AMD.

    2) There’s no sleep mode for a business Even if you wanted to, it is usually a very, very bad idea to put an operating business into sleep mode. The main reason is that a lot of the expenses associated with that business continue whether you do anything or not, with no revenue to help pay for it.

    In the case of AMD, the biggest example of this is that the people who lent you a couple billion dollars to build the fab want their interest (and eventually the money they lent) back

    Working or not, AMD still has to spend a ton of money converting Fab36 to 45nm. Again, a lot harder to do if you stop collecting a paycheck.

    Even expenses you think you could cut, well, you can’t. You can’t lay off all the fab workers because the German government who lent/gave you a lot of money made job guarantees for those workers part of the package. Even if you could lay them off, eventually you’d have to rehire a staff, and that’s not like staffing a McDonald’s.

    AMD would lose much more money if they stopped making CPUs than if they kept going.

    Even if the writer just meant stopping current K10 production, well, there’s not much to stop, and whatever’s been already spent is spilled milk. Making K10s may not be too good financially, but it’s definitely less bad than doing nothing, and probably less bad than making additional X2s.

    3) CPUs Aren’t Built In A Year A truly new CPU architecture doesn’t go from start to finish in a year; five is more like it. Even under Intel’s “tick-tock” system, a lot of preliminary work is done before a new architecture becomes official.

    Even if you cut some slack on what is a “new architecture,” to “an old architecture with some big changes,” the process still takes two-three years, Nehalem is a good example of this.

    Show me a “new” CPU completed in 1-2 years, and I’ll show you a tweaked CPU, period, and tweaks don’t get big performance improvements.

    It will take AMD a number of years before they can do anything that would be a serious change from the Hammer architecture.

    Fortunately/unfortunately for AMD, architecture isn’t the K10s foremost problem: speed is. A K10/K10.5 able to run at the same speed (deault and overclocked) as a Penryn/Nehalem might still come in second, but it would be close enough to get AMD out of the bargain basement and make some money.

    So long as CPU price is mostly determined by speed, AMD has to get faster to get out of its fiscal bind, and even being optimistic, it doesn’t look like they can do it in 2008. Maybe matters will look better if AMD comes out with a high-K 45nm 2.0 version; they probably will look better at 32nm.

    But how much time does this company have to fix these fundamental problems?

    When you look at the issue of AMD’s survival, there’s really two questions. The first is “Can the patient last the night/week/month/year?” and the other is “Can this patient become healthy again?”

    If you look at what the company is doing (as opposed to what they’re saying), they’re retrenching and stabilizing the patient.

    Can they keep the patient from flatlining the next year or two? Probably (though a serious recession drops the likelihood of that). Can they get the patient back to robust health? Probably not. It looks like only outside intervention in the form of a chained Intel will do that.

    Ed


  • Leave a Reply

    Your email address will not be published. Required fields are marked *