We’re now getting a number of reports about Barcelona’s actually showing up in about three months, but in the hubbub of all the activity, two points need to be made:
I’ve said it before and I’ll say it again; server and other high-end chips are a fringe benefit to a CPU company; they don’t pay the bills.
AMD isn’t losing lots of money because of their server chips; they’re losing lots of money because they can’t sell a lot of chips and sell them profitably.
While the desktop is becoming less important as a share of overall CPU sales, what is taking its place is notebook chips, and while the official AMD prices look decent, any look at the prices of actual AMD notebooks indicates that those chips are being sold at a steep discount, too.
Server and “enthusiast” chips can only slow the cash burn rate; they can’t make AMD profitable all by themselves.
AMD just isn’t going to sell a whole lot of these chips in 2007, probably somewhere in the lowish hundreds of thousands. Quad-core isn’t going to become mainstream until the prices get much lower than they are today. Now you may be happy to pay $270 for a quad shortly, but your Sixpack buddy won’t be. Quads will probably not become mainstream until late in 2008 on the desktop, and the question isn’t when but if (at least at 45nm) for notebooks.
So dual-core is going to be the mainstream for the next twelve-eighteen months, and if AMD was ready to make lots of fast-enough duallies, that’s what they should be doing.
They aren’t, so it must be presumed that AMD can’t make K10 mainstream yet.
65nm SOI: The Real Problem?
What’s been overlooked in all the Barcelona babble is that AMD still hasn’t gotten the 65nm process right yet.
Almost eight months after introduction, the fastest AMD chips you can buy today are still 90nm CPUs. Until the other day, the fastest 65nm AMD chip clocked in at just 2.6GHz, now it’s . . . 2.7GHz (and there’s few 65nm 2.6s around).
All Opterons are still 90nm. Outside of maybe the mysterious TK-53 which has shown up the last two weeks (and which runs at just 1.7GHz), all Turions are 90nm, and AMD recently promised that they’d show up by the end of June.
Does this sound like AMD has 65nm down pat? I think not.
The K10 generation may well have other problems, but if AMD can’t gun up the speeds on one process shrink or duplicate it on others, is it surprising that they’re having even more difficulties with a somewhat new design?
If first-generation 65nm chips are still a relatively small part of AMD production, should it be surprising that AMD would be constrained building second-generation chips?
That may not be the only problem K10 has, but I think it’s pretty safe to say it’s an awfully big one on the list.