The Costs of Dual Core

The Inquirer has cadged Mercury Research’s marketshare results from the last quarter.

The numbers were quite good: 21.4% of x86 production, but what is interesting about these numbers is that they indirectly seem to indicate AMD’s current capacity.

In the third quarter, based on AMD’s statements, it seemed that AMD made a bit more than 9.5 million CPUs. However, Mercury’s marketshare figure indicated that only about 9 million found themselves a home in a PC that quarter.

In the fourth quarter, again based on AMD’s statements, it seemed that they made somewhat more than 12 million CPUs, but the PC figures would indicate that somewhere close to 13 million PC had green inside.

There’s nothing fishy about this, the differences can easily be explained by the time lag between finishing a CPU and selling it in a PC box, and a little stockpiling in the third quarter in anticipation of the Christmas quarter, but it does seem to indicate that Fab 30 is working flat-out.

1+1 = 1?

AMD probably can’t put out more than, say, 12.5 million CPUs a quarter from Fab 30. They might improve marketshare a bit more simply because fewer computers get sold the first quarter of a year than the preceding quarter.

No big deal, many would say, Fab 36 will be on line soon enough to provide more capacity.

But just how much? Remember than Fab 36 was supposed to a 65nm fab from the start, and it isn’t going to be a 65nm fab from the start because they couldn’t get certain equipment needed to mass produce 65nm chips, so they have to start at 90nm.

Granted, much if not most of the fabbing equipment being installed today in Fab 36 can make both 65nm and 90nm chips, but what about the equipment that isn’t available for 65nm. AMD has to install something else until they can buy the new stuff. Where is that coming from? Are they buying additional 90nm fab equipment to use for just a little while? If so, how much is that going to cost them? Or are they just moving some equipment from Fab 30 to Fab 36 for a little while? That seems to be borrowing from Peter to pay Paul.

Moving further down the timeline, AMD expects to have capacity for about 90 million CPUs by 2008-9 (not including outsourced CPUs from Chartered). Assuming AMD is at least at 65nm by then (in all likelihood, 45nm migration will occur during 2008), at first glance, that’s not a lot of CPUs to get out of two working fabs.

We can solve that mystery pretty easily, by assuming that by 2008-9, just about all CPUs will be dual-core. Two cores mean half the CPUs produced. Dual-core (along with additions like cache) means that any CPU company making them is essentially forfeiting a generation’s worth of increased production from a process shrink.

Dual cores mean AMD’s forecasts of 90 million CPUs inhouse make sense. It also means AMD can’t hope to overtake Intel, or even become Pepsi to Intel’s Coke in the next few years.

Had the world stayed with single cores, two 65nm fabs at Dresden could have pumped out over half of expected world demand for CPUs a few years from now. With dual cores, the figure is more like 30%.

That’s not entirely good news for Intel, either. Dual cores mean half the CPUs for them, too, especially with all the cache Intel has been adding. For them, dual core means building more fabs than they would otherwise, and that’s a strain even on Intel.

Pretty expensive blue crystals, aren’t they? 🙂


Be the first to comment

Leave a Reply