Important Athlon64 Info

Add Your Comments

X86-Secret provides some leaked AMD documents that finally provide some news you can use in figuring out when you might want to go to Hammer (I wouldn’t bet they’ll be up very long, so look quickly). The comments are in French, but the charts are in English.

To hit the high points:

  • There will be three 130nm “big” Athlon64s: the 3100+, 3400+ and (in early 2004) the 3700+. They will be expensive to very expensive.
  • 90 nm Athlons will come roughly a year from now. They’ll start at 4000+, with a 4300+ due a quarter later. They will be very expensive, too.
  • A 64-bit Windows will reach release candidate status around the time of product introduction, but the final version won’t be released until the end of the year.

    Overclocker Analysis

    The 130nm big Athlon64s will definitely get up to a real MHz of 2.2GHz (there are hints the A64 Jrs might get up to 2.4GHz). This is about what current Athlons are doing with default voltage, so this should only grossly disappoint AMDroids with unrealistic expectations about SOI.

    Overclocking the original A64s will likely require voltage modification; based on these documents, these CPUs look like they’ll be running at very close to the maximum voltage range (1.5V, with a 1.55V maximum).

    A much bigger discouragement to overclockers should be the pricing. Pricing on the initial Athlon64s looks like it will exceed $400 initially, and not move downward too quickly. The pricing on the cheapest big A64 may dip below $200 in the first quarter of 2004 (but see below).

    The 130nm “big” Athlons aren’t going to be around very long. Last orders for the 3100+ are supposed to be first quarter 2004, for the 3400+, second quarter.

    This might prove to be a blessing in disguise for overclockers. It would be foolish for AMD to stop making chips for these speed grades, but 130nm CPUs with 1Mb cache are expensive to make. What is far more likely to happen (but see below) is that AMD will introduce 90nm versions of these chips (which should cost them less to make) when or shortly after the faster 90nm “big” Hammers show up.

    Those will be the CPUs that will likely be the overclocker’s chips of choice. They’ll probably cost $150 or less, and at 90nm, should provide pretty good overclocking room.

    Expect them in about a year if all goes well.

    Or Maybe This Is Something Quite Different

    What I just said above presumes business as usual.

    If you look at the roadmap, though, it indicates something quite different.

    So far, we’ve just talked about “big” A64s, those with 1Mb of L2 cache.

    But that’s not the only A64 that will be made. There will also be an A64 made that has only 256Kb cache. Call it A64 Jr.

    The roadmaps are very odd about this chip. With one exception (which may be a typo), these chips don’t show up at all in the system pricing roadmaps.

    The exception is the pricing of the 3100+ chip for 1Q 2004. The notes on the 3100+ for earlier quarters indicate that it is an A64 Sr., but for the 1Q, it says that it is an A64 Jr.

    However, the A64 Jrs show up all over the place on the longevity roadmaps (including the substitution of 90nm chips at lower speeds as described above; there’s no such substitution shown for the “big” A64s). 90nm A64 Jrs. show up at the same time as their big brothers, or a quarter later, at both low and high speed grades.

    Perhaps most importantly, there is very little gap in PR rating between the two, certainly nothing like the rating gap that’s existed between the “senior” CPUs and Durons/Celerons.

    Does AMD intend to have the A64 Jrs. as the “regular” Hammer desktop CPU, with the big A64s (which may well slip over into “Opteron” status being super-premium chips with super-premium prices? That would also explain why the lower-speed A64 Srs. get pulled quickly: this class of processor just won’t be sold cheaply.

    How would you feel about that?

    Second-Class Citizenship?

    This might not be a bad idea for AMD selling to the general public, but what about you?

    If this is what AMD’s future is going to be (and heaven knows they could change their plans three times in the next year), at least some AMDers are going to face some ugly choices.

    It’s very hard to see how what is basically going to be a souped-up Duron is going to beat a Prescott. The difference in cache will likely give the 1 Mb Prescott about an 8-10% advantage over a 256K A64 Jr. Add in other differences like dual vs. single-channel DDR (all A64 Jrs. will be single-channel) and hyperthreading (which should be used more often and effectively by apps and games a year from now), and the gap could end up anywhere between 15-30% (probably closer to 15 than 30).

    Or you can buy an A64 Sr. at Opteron-like prices for a fighting chance to match.

    What would you do under those circumstances?

    Additionally, it is likely that those PR ratings will only approximate Intel equivalents while using a 64-bit OS. Unless your answer to OS questions is, “Warez my copy?,” this will represent an additional cost to upgrading AMDers.

    On the other hand, such systems are likely to be a good deal cheaper than Intel systems. In today’s world, will that matter more?

    Should this be the future, then again about a year from now, people will be taking these 3100+ A64 Jrs. and cranking them up (probably quite a bit), but they’ll probably not be able to compete against Prescotts.

    AMDers seem happy enough doing that now. Will that remain true if the same is true for desktop Hammer-systems?

    Which Way Will They Go?

    Your answer to that question really depends on what’s more important to you: relative performance or relative cost.

    If you want to have the fastest not-too-expensive computer around, this is likely to be bad news for you, and if this scenario happens, you’re likely to defect to Intel.

    If you want to have the fastest very cheap computer around, this isn’t likely to perturb you very much.

    How many AMDers are behind door number one, and how many are behind door number two?

    My Best Guess

    Should the “Duron for Everyone” strategy actually go into effect, my best guess is that those who are more interested in performance than cost will end up going to Intel. This will be a significant but not wholescale defection.

    However, the majority of current AMDers will stick with AMD so long as the price is right.

    The real danger for AMD will be if they try to keep up the price of the A64 Jrs, too. Then they’ll lose both groups.

    Don’t Decide Now

    The late spring/early summer of 2004 will be an important decision point.

    Not only are 90nm Hammers (big or little) supposed to show up, but so will Intel socket 775 mobos and processors (Prescott-class processors should be affordable by this time, also). PCI Express will make its debut in mobos around this time, also.

    Even if the thought of your only effective AMD option being an A64 Jr. turns your stomach, your reaction shouldn’t be to run out and buy a PIV system. First, it’s not particularly cost-effective. Second, even if I’m right and this is what AMD plans to do today (which is hardly a certainty), I wouldn’t bet that they’ll stick to it.

    No, what you should get out of this article is that about a year from now will begin the period when you ought to start making your socket A upgrade decision, based on the situation then.

    You may well decide to wait longer than that, but it will be a year until you’ll have upgrade choices on both sides that will give you a significant advantage over what you have/can get now, and justify a major upgrade.

    If you’re hurting now, you might as well upgrade now, waiting on current equipment won’t save you much or get you much better. After now, the next big window begins around June 2004.

    What Do You Think?

    This isn’t a poll, but if you have any opinion on whether or not you’d stay or go should the regular A64 chip be the “Junior” model, you know where to find me.

    Ed

  • Leave a Reply

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