A More Benign Explanation? . . .

Add Your Comments

People are suspicious about just how the new AMD motherboards manage to overclock more. We make an educated guess as to what is really happening, and suggest the coverup is worse than the crime.

Anandtech has been looking into a motherboard that includes the new SB750 south bridge, which AMD claims will let their Phenoms overclock more.

They report that:

  • Yes, it works, to varying degrees.
  • It seems to work better on “lousy” processors than better ones.
  • AMD calls the feature Advanced Clock Calibration, but doesn’t want to explain what it is or what it actually does. This leaves the Anandtech testers more than a little suspicious, and openly wondering if AMD is pushing the envelope too much and compromising the integrity of its processors.

    A More Benign Explanation?

    Let’s get a couple minor details out of the way first. Motherboards with the SB 750 have a six-pin link directly to the Phenom chip. Six pins can’t do much of anything, so it’s not like this is some mini-Hypertransport link. It looks more like a supplement to what a BIOS would normally do.

    Something we know this does do is open up an additional multiplier for a Phenom. To make a long story short, doing that is probably the occupation of one, at most two of those six pins.

    What are the rest of those pins doing? It’s very likely they all have something to do with this ACC, but that doesn’t tell us what it is. Knowing that four or five pins control everything does tell us it can only be doing one or a couple things.

    What could that be? Again, let’s review what we know:

  • It helps overclocking.
  • It helps lousy CPUs more than good ones.
  • When AMD doesn’t want to talk about something, there’s something embarrassing they’re trying to hide.

    Is there an explanation that fits all the facts? I think so. Now mind you, just because an explanation fits all the facts doesn’t mean it has to be right. That could just be coincidence, and I could be barking up the wrong tree.

    This is an educated guess as what it going on, and while I may be barking up the wrong tree, I’m pretty sure I have the right part of the forest.

    1) Even though they shouldn’t on paper, different CPUs differ in performance, some are better, some are worse. Over the course of time, with an odd fact here and there, there’s been signs that AMD processors vary a bit more than Intel’s in their electrical properties. That could be due to a less-than-ideally consistent manufacturing process or just a more lenient vetting process.

    2) CPUs fail in overclocks usually due to timing issues. If you thought it was heat, well, a lot of heat changes the electrical properties of a CPU, and that causes the timing problems.

    3) CPUs do not dynamically adjust timings to fit an environment. They’re basically designed to work within a certain frequency and temperature range where they can reliably meet timing requirements.

    4) There are certain manual adjustments that can be made in the BIOS of certain mobos which can serve to adjust these timings and allow a CPU to run at a higher speed than it could otherwise. This is very similiar to loosening the timing on memory. These adjustments go under the name “clock skew,” and you can read some more about one form of it here, and see an example of clock skew controls here.

    5) This ACC apparently has 24 different settings, which would take five pins to implement.

    Taken all together, saying that ACC is just another way of saying “clock skew” is at least a plausible explanation for what ACC actually is. It is an established overclocking tweak, and it would tend to help poor processors more than fast ones. Calling an established tweak something new is definitely embarrassing enough for AMD to obscure, and less obviously, the implication that some Phenoms need a little remedial help isn’t something to brag about.

    But even assuming all this is 100% true, this is hardly very bad from a technical standpoint, and you couldn’t say it compromises the integrity of the CPU. What’s happening here? Some timings are being relaxed by some miniscule amount. That’s hardly a mortal sin in overclocking; it’s more like a requirement.

    OK, no full marks for honesty here, but this is stupid dishonesty. If AMD has just called it what it was, a few hypertechies would have complained, but the average person actually using it would have considered the additional speed made possible by AMD popularizing this lesser-known tweak a lot more important than the exact whys and wherefores.

    Yes, old habits die hard, but here’s an great example of how slickness blows up in your face. Assuming I’m more or less right, AMD is covering up what at most is a minor embarrassment, but the covered-up not only know there’s a cover-up, they’re saying they think it’s a lot worse than it really is.

     

    Ed

  • Leave a Reply

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