Socket 939: DOA

What About Socket 939?

All you need to know is that AMD doesn’t want you to buy one anytime soon. The price for the entry-level socket 939 is even more ludicrous than feared, $500. AMD apparently is going to charge about a $100 premium for that extra 100 PR points over the Intel standard. Not completely unexpected, but bad news nonetheless.

If you want a dual-channel system at a reasonable price today, AMD has a very simple solution for you. Buy Intel.

That’s what makes this positioning and pricing so absurd. AMD is effectively treating dual-channel as a $300 extra option, while it comes standard with Intel.

The problem is not charging $500 for a 3500+. The problem is not having little brothers to the 3500+ while Intel does. Whether it’s socket 478 or the upcoming socket T, Intel has/will have mainstream-priced dual-channel processors for it.

Performance is about as expected, a dual-channel Newcastle generally (but not always) does a little better than a single-channel Clawhammer. The socket 939 mobos don’t seem to be terribly well optimized yet, but doing so will only add a few percentage points to performance.

The high-end processors don’t overclock very much, but that’s to be expected at the tail-end of a process generation. Getting a Hammer to go at 2.55-2.6GHz at 130nm without extreme measures is pretty good work by AMD.

A Little Math

For those who object to what I say about the positioning of socket 939, saying, “AMD needs to make money!” I quite agree. This isn’t the way to do it, and I’ll prove it right now.

According to some projections, AMD is supposed to sell 50,000 socket 939s this quarter. Let’s assume the ASP on these is $600, and it costs AMD $50 (probably too high, but I want to be conservative about this) to make one

You get:


Let’s assume AMD launched socket 939s across the line, down to the $170 mark, and they sold 200,000 socket 939s instead. Assume an ASP of $200, and the same direct cost of $50.


Isn’t this amazing? By charging less and getting more volume, AMD actually makes more profit.

Actually, they’d probably sell rather more than 200,000 systems. Let’s run the numbers at 350,000.


Almost double the profit!

Granted, this is rather oversimplified and doesn’t take into account the effect on socket 754 or socket A sales, but I think it very safe to say that such pricing (which as you can see is still very profitable for AMD, certainly more so than socket A sales) would increase overall demand for AMD products. For instance, some of that increased volume would come from people who would otherwise buy from Intel.

But no, somebody thinks that if they charge high enough long enough, you’ll start wanting to pay more for AMD products than Intel. Even when the features you say justify the price premium are either already or are going to be offered by Intel at no or very little extra cost.

The One Thing You Need To Remember

Eventually, AMD is going to have to sell more than a handful of these things. There is no way in hell they’re going to be able to sell millions of these things at $500, or even half that price.

If You’re Fat and Happy

Those of you with fairly current socket A systems are generally well-satisfied with them and are perfectly willing to wait AMD out until:

a) Hammers offer a big performance jump over socket A systems and/or
b) AMD comes to its senses on pricing/positioning.

This has and continues to make much sense.

But I’m Lean And Starving!!…

But I’m Lean And Starving!!

It’s easy to pass up a meal when you’re not hungry.

However, not everyone is in that position. There are those who upgrade infrequently and who are hurting now.

They have two choices:

  • Go with a socket A system as a stopgap for a while or
  • Settle for socket 754

    The Real News for Overclockers

    The real news that has been emerging recently is not about socket 939, but socket 754. Recent events have made it a better choice than it was even a month ago.


    For The Present

    PCI/AGP Lock for Mobos Without a PCI/AGP lock, the mobo became the bottleneck to Hammer overclocking, especially for slower Hammers like the 2800+.

    Now that nVidia AND Via have implemented PCI/AGP locks on their latest generation of mobos, the mobo stops being a bottleneck.

    CG Stepping For Hammers The latest stepping for Hammers offer a modest increase in maximum performance, but every little bit helps. It’s probably reasonable to assume that a 1.8GHz CG-stepping Hammer ought to be capable of 2.4-2.5GHz without much fuss.

    For The Future

    The next serious performance jump ought to occur about a year from now, when 90nm Hammers become affordable. After that, the next big jump will be when dual-core something (whether Intel or AMD) becomes affordable, which will probably be two years from now (and general dual-core support will probably come even later that that).

    Looking at the first, even if you make generous assumptions about 90nm overclocking, the performance jump will probably only be about 20-25%. This is a bit less than the current gap between socket A and Hammers today.

    That jump will likely be enough to get most of the fat-and-happy socket Aers moving, but is it worth a year of hurt if you’re already hurting?

    Looking at the second, for those hurting now, buying now puts those who do in a position to leisurely upgrade to dual-core when its time comes. Those who wait for 90nm may feel a little short-stepped.

    Another positive factor is the near-certainty that x86-64 will become generally adopted due to Intel’s grudging moves to include it, though it probably will be towards the end well into 2005 before that has a big impact.

    One Skeleton Still In the Closet

    Yes, any socket 754 buyer is still staring at instant obsolescence, but anybody buying ANY system these days is going to have a relatively short lifespan before his platform gets superceded. For the next couple years, looks like every system will end up dead-end sooner rather than later. Also keep in mind that the candidate for a socket 754 today isn’t a frequent upgrader, so this is less of a factor for such people than it is for others.

    For those who cringe at a $175 price tag, there will always be the Paris chips in four-six months, but it’s probably not worth it for the lean and starving to go without for another six months and forfeit a now more-viable x86-64 just to save maybe $100.

    Hungry? Then Eat. Not? Then Don’t

    If you’re computing guts is empty and hurting, a CG-stepping processor combined with a mobo with PCI/AGP lock ought to relieve the hunger pangs. If you want to be able to figure out what CPUs for sale are actually CGs, see this.

    I checked Newegg’s pictures of the lower speed Hammers, and found that the desktop CPUs are still the old stepping, but the mobile processors have the new one (AX is the usual CG designations for mobiles). That doesn’t necessarily mean they’ll work in desktop mobos, so beware.

    What all this flurry of activity shouldn’t do, though, is get you hungry. If you’re fat and happy with socket A, this is no more than a snack. If you want a full meal the next time you gorge, wait.


  • Be the first to comment

    Leave a Reply