AMD Thinks It Will Make A Profit This Quarter
Given the upsurge in CPU sales the last few months, and an additional boost due to initial desktop Hammer sales, this would hardly be surprising, as we’ve indicated a few times in the last couple months.
The issue is: can AMD keep it up throughout 2004?
90nm Hammers To Be Delayed A Little
A matter of rather greater concern to this audience is when the second generation of Hammers is due out. A roadmap from x86-secrets.com shows the schedule for 90nm Hammers has slipped from the first half of 2004 to the second half of 2004, and AMD confirmed a delay in their recent analyst meeting.
However, the comments at the analyst meeting seemed to indicate initial 90nm production shipments of what we know now as FXs late next summer, with other models to follow.
This trend is reinforced by the introduction of new 130nm processors in the second quarter of 2004. Unless AMD plans to introduce these, then introduce 90nm versions almost immediately, it’s probably safe to say that there will be a significant amount of time between the two events.
That may be a good thing. If you think you’re confused now, wait until next year.
Too Many Plums In The Pudding?
First, there was 940-pin Opteron.
Then came the socket 940 Opteron That Isn’t An Opteron (aka FX) along with the Opteron That Really Isn’t An Opteron (aka socket 754 Athlon64.
Next year, the flood.
First, we’ll have another socket, socket 939. There will be two kinds of 130nm processors for this socket: one with 1Mb of cache, the other with 512K of cache. The latter will be called “Newcastle.”
Newcastle will also be made in a socket 754 version.
This means that by second quarter, 2004, AMD will be selling the following:
Not enough? Be patient.
Next, we’ll get 90nm versions of those socket 939s. The 1Mb cache version is being called San Diego, the 512K version will be called Winchester.
So by fall, we should have seven versions of desktop Hammer. Well, no, actually not. We’ll have a 90nm version of socket 940 at this time, too (that’s called “Venus”). So we probably have eight versions for sale next fall.
Some companies might think that’s enough. Not AMD.
Shortly after that, just in case you’re allergic of x86-64 or 90nm, you’ll get “Paris.” According to the AMD roadmap used at the analyst conference, that really is supposed to be a descendant of Athlon XP. Paris is supposed to be a 130nm SOI version of the Athlon XP, to be followed a year later by a 90nm version called “Palermo.”
So next Christmas quarter, it looks like AMD will have eight different types of desktop Hammers floating around for sale using three different sockets (plus a revamped Athlon XP which might outsell them all):
And for each of these three different sockets, there will likely be two types of mobos for it: the current AGP/DDR/PCI generation of systems, and the future PCI Express/DDR2 generation of mobos.
And, oh, BTW, in all likelihood, it’s by no means certain that all CPUs of a certain socket type will work right in all mobos of the same socket type (it’s that danged memory controller and whether it handles DDR, DDR-2 or maybe both).
Are we confused yet?
Trying To Make Sense of It All
It looks like AMD will try to sell FXs and A64s at a premium price the first half of the year, and only start to make them affordable the second half of the year.
After the 130nm-90nm and mobo intergenerational conflicts end, AMD looks like it will have four Hammer lines, which seems to be at least one or perhaps two too many. There will be a high and low-end FX, and a high and low-end A64 line. There will also be AthlonXPs to essentially fill the Duron slot for at least a couple years to come (and which may very well prove very stiff competition against the newer lines).
AMD seems to want to make FXs a superpremium product line, A64s the mainstream line, and XPs the discount line. However, with the multiple product lines, you end up with “low” FXs and “high” A64s. It is also questionable whether a single-channel memory architecture is going to be terribly desirable or even viable in the long run for another two years or more other than as a value product (in which case, you don’t need socket A).
You can make a number of different arguments on what product lines ought to go, but it’s hard to justify all of them.
Snatching Defeat From The Jaws of Victory?
It looks like AMD finally ought to have a competitive product throughout 2004, but it’s not enough to have good players. You need good coaches and a good gameplan, too, and frankly, this looks dazed and confused.
There’s too many products across too many product lines which will end up competing against one another (for instance, which will be better, an FX with 512K cache, or an A64 with 1Mb cache?), and odds are that if the product positioning is right, the price won’t be.
Frankly, I look at all this, throw up my hands, and say to myself, “Get your act together and then talk to me.”
Do you know what I (and I think a lot of you) want, say a year from now? I’d like a 90nm socket 939 dual-channel DDR2/PCI Express system that will cost me about the same as a PIV system would cost me today.
Unless Prescott really goes to hell, I know I’ll be able to get something like that from Intel a year or less from now. I greatly suspect that AMD could easily give me (and you) something better than Intel for the same price or less roughly around the same time, but I don’t think they will.
It seems like AMD is intent on offering me everything under the sun except what I want. I don’t want a socket 754 box for a performance solution, a single-channel solution is an anchor. I don’t want to pay an arm and a leg more just to get dual-channel capacity.
If those are the choices AMD gives me, well, there’s always that other company.
The Wrong Focus
During the analyst conference, AMD had a bizarre graphic which illustrates the core problem I think the company has.
It shows two pyramids. The first pyramid shows AMD’s earlier focus, with the base being the desktop, with mobile and server sales being addons. This makes a great deal of sense since desktop processor sales are AMD’s daily bread, while the others are butter. Intel certainly knows that.
The second pyramid shows the new “improved” AMD focus, with an inverted triangle showing servers as being the top priority and the desktop being given lowest priority.
This is the kind of mindset that killed IBM’s chances of monopolizing the desktop. You can try to get more butter on the bread, but you can’t just make butter. If you do, you’ll starve to death from lack of nutrients.
The server market isn’t and never will be big enough to pay the fab bills, and it will be years before AMD can hope to be a major player in the server market (which is only about 5% of the desktop market). (Mobile could well eventually replace desktop as the main man, but not for a while yet.)
In the meantime, you can’t cripple your desktop offerings (either feature or price-wise) to protect your higher-end server offerings. That will get you an inverted pyramid, alright.
Have you ever tried to keep an inverted pyramid standing up?
If you’ve wondered why I’ve been so down on AMD, even after they managed to get Hammer right, this is the core reason. The players now look good enough. The coaches and gameplan don’t.