There’s a report from Goldman Sachs floating around about AMD. A good-sized chunk of it can be found here.
Anything new? Not especially, outside of a few tidbits.
AMD is now beginning to ship notebook 90nm CPUs, with desktop chips to follow in a month. That means notebook machines in mid- to late-September, desktops in October.
It’s not surprising that notebooks come first. The story on current Athlon64 machines has been “fast machines, even faster power consumption.” This is no surprise given that most of these notebooks use a 62W A64 chip (the low-power 35W version up to now has only been found in high-end A64 machines), which doesn’t compare too well with Dothan’s 21W.
While 90nm A64 chips probably won’t get down to 21W, they’ll get close enough to make relative battery life a negligible factor.
For desktops, the issue hasn’t been if AMD could get a 90nm CPU out the door, but rather how fast they can make it. IBM had little problem making a 2Ghz 90nm SOI chip; they had a much rougher time making a 2.5GHz chip.
Right now, the evidence indicates that AMD can make 2.2GHz 90nm chips (engineering samples exist and have been tested), but not 2.6Ghz chips (making 130nm 2.6Ghz CPUs makes no sense if you’re confident you can make them at that speed at 90nm).
It’s possible that there will be a last minute breakthrough, but until we see evidence to the contrary (i.e., an AMD statement or a leaked engineering sample), we think it more likely than not that the first desktop 90nm chips will debut at speeds of less than 2.6GHz, and probably won’t overclock very much. We expect to see them debut in the middle of the product spectrum, and migrate downward sooner rather than later.
One interesting item in the report is that AMD mentioned the relative ASPs of Hammers as opposed to Athlon XPs. Hammers have three times the ASPs. Estimate the actual dollar figures, and you get an ASP of $180-200 for Hammers and a low $60-ish figure for Athlon XPs.
The Hammer ASP figure is interesting, not because it’s so high, but because it’s so low compared to retail prices of most Hammers. While CPU ASPs always lean to the bottom end, the Hammer figures are especially skewed in that direction. This indicates that sky-high pricing has done very little for AMD.
The report also notes that Hammer ASPs will decline in 2005, “as AMD drives AMD64 into the lower end (our emphasis) and broadens its reach.”
This is a confusing statement for the more knowledgable. Does this mean Athlon 64s will drop a lot in price, or are AMD64-less socket 754 Semprons being included in this picture? Does that imply that socket 754 Semprons will eventually get x86-64 activated? Or did the analyst know not the implications of his words?
Ww would bet on the last. While we think Semprons will eventually and inevitably activate x86-64 on Semprons, we think Intel will end up deciding when that will happen when it does the same with its Celeron line.
Will Socket A Go Away On Schedule?
Another decent-sized tidbit is an AMD estimate as to how many Hammers they’ll make in 2005. They “hope” to hit 20 million. Given that they won’t make all that in Q1, and maybe 4 million in Q2, to reach close to twenty million, that would mean the vast majority of AMD processor will be Hammers in Q3 and practically all in Q4.
What that does to their low-end socket A Semrpon strategy is beyond me, seems like another six-month strategy to me. However, AMD has always had a bit of a problem getting people to give up an old standard instantly. Those who have been around awhile will recall that the K6 took much longer to die than AMD either wanted or expected, and AMD even shrunk the last K6s to 180nm to make them a bit faster.
Will history repeat itself? Will there be a farewell 90nm socket A, say, towards the end of next year?
Share and Share Alike
For servers, AMD repeated its message that it has a 4% server market share now, and hopes to get it to 10% by year’s end. Nothing new there.
What is new is AMD’s expectations for 2005. It hopes its market share will continue to grow “ultimately toward 20%” (talk about vague wording; going from 10% to 11% would do that) and that Dell will jump in once their market share approaches 20%.
What’s surprising about the statement is not what AMD believes about Dell (it’s probably a pretty good assessment), but that the oh-so-secretive AMD would actually say so in public.
That ought to make for some interesting future negotiations between Dell and AMD, as well as talks with Dell and the Blue Men.
Unfortunately, the analyst who wrote this either hasn’t been following AMD very long, or doesn’t know what he needs to know.
For instance, 90nm Hammers aren’t arriving on time; they’re arriving a year late, much like Prescotts. AMD has had plenty of problems getting 90nm SOI on track, as its past payments to IBM (which, BTW are continuing, though presumably more for future 65 and 45nm production these days) to help them with SOI indicate. He doesn’t apparently know this, and just reports what he’s been told.
More importantly for now, AMD’s relative competitiveness and profitability over the next twelve-eighteen months are going to depend very heavily on how well they can ramp 90nm SOI. If AMD can crank out 3GHz+ Hammers whenever they feel like it, they could do all sorts of things to Intel that they can’t if they’re struggling to get 2.6GHz out of them.
Yes, it’s a techie detail, but it’s the critical techie detail any financial analyst needs to know to get a good answer to the very basic question, “Do you have a better product than the competition, and if so, by how much?”
Obviously, it’s not the only important detail, but it’s a biggie.
I don’t want to especially pick on this particular person, because over the years, I’ve never seen any CPU analyst for a major stock firm who has shown any real comprehension of these issues.
Obviously, an analyst shouldn’t be asking AMD execs, “How well will your new Hammers overclock?” However, if they’re telling people what to do with their money, they damn well ought to be able to ask, “Just how much better is your product going to be than Intel’s?” then ask for and comprehend the hard cold technical details.
An Assessment From This Angle?
Despite Intel’s huge problems, AMD will only manage to do little better than muddle through the next year or so. Yes, there will be some improvement, maybe they’ll make an extra $50 or maybe $100 million from CPUs for a couple quarters in 2005 (indeed, AMD’s flash memory business may do rather better), but that will be about it. No radical improvement, no big shift in the relative correlation of forces.
Intel will beat AMD to 65nm in late 2005, which will put AMD back on the defensive until they can match with Dresden II, realistically in the second half of 2006.
Thus, the technical advantages of Hammer will not lead to any more than transitory financial advantages. Perhaps AMD will be left with a somewhat larger share of the server market, but keep in mind that a 10% share is still rather less of a share than it has for the desktop market.
The core problem is that SOI isn’t giving or is going to give AMD all the advantages they had planned on getting, and not enough to give it a decided advantage over Intel, probably not even to the degree AMD had over Intel in the Thunderbird/PIII area.
I also suspect AMD has known this for quite some time and have shifted their plans accordingly.
That doesn’t mean SOI was a mistake; AMD would likely be in deep, serious trouble given its architecture if they hadn’t.
But this is not an updated version of Jack and the Beanstalk. It’s more like “Staying even (or getting a little ahead) or the Joneses.”