AMD lost money again last quarter, but that was solely due to their flash memory division doing horribly.
Profits in the CPU industry went up a little bit, but this is something that ought to happen as production shifts from lower-priced XPs to higher-priced Hammers. A bit over a third of AMD’s production consisted of Hammers, but they brought in 63% of the revenues.
So unless Hammer prices go into the toilet, the profits from AMD’s CPU division should increase sizably the next few quarters.
AMD also said it planned to partially spin-off the flash memory business. All we’ll say about that is the best time to do something like that is not when the business you’re spinning off is losing a ton of money.
Now that we’ve gotten the boring finances out of the way, let’s talk about the items far more likely to interest those reading this.
Semprons Will Get x86-64 We have long said AMD was going to have to do this given Intel’s decision to make Celerons x86-64 compatible, but AMD finally said so officially during the conference call. When asked if they would do so by the end of the year, someone from AMD said, “Most likely.”
AMD64 compatible Semprons will probably show up long before then, probably shortly after Intel gets x86-64 Celerons out.
DDR2 Will Arrive In 2006 Though there’s been more hints about this, AMD finally said officially that they “have a planned transition to DDR2 in 2006.” No comments on what processors would or wouldn’t work with DDR2, but it would probably be very safe to assume that current Hammers will not, and anything that will work with DDR2 will be some socket 1200+.
Those are the two hard news items.
Now, The Fuzzier Stuff
Unfortunately, AMD is never very forthcoming in these conference calls. They always refuse to answer a lot of questions, particularly those that involves numbers, like “How many socket 64s did you sell last quarter?” and especially those questions with numerical answers which might take the shine of whatever spin they’re giving.
This leaves people trying to figure them out attempting to come up with more indirect means of determining what is going on.
In the past, a very good tool for deciding how well or badly a particular AMD project is doing is how willing or unwilling the AMD execs are to talk about them.
Let’s see who’s the crazy aunt in the AMD attic this time around.
This time around, the major candidate was desktop dual-core chips. AMD was plenty enthusiastic about server dual-cores, and flat out said they planned on charging a lot for them, but the enthusiasm fell right off the table when it came to the desktop. The message was “We’re going to wait until the software’s ready.” They talked a lot about Smithfields not being “real” dual-cores, and said that they’re single processors would be better than any Intel dually well into 2006.
Take that literally, and we won’t see AMD desktop dual-cores for another year or two. They aren’t going to do that, they said they’d make some by the end of the year (long before we’ll see any big shift to multithreading), but they sure didn’t seem to terribly want to.
I don’t think we’re going to see desktop dual cores anytime soon, and when we do, there won’t be very many. Let me correct that, there won’t be many (or any) at the low-ball prices Intel plans on charging for the low-end Smithfields.
AMD doesn’t have a feature they can take in and out of processors that at least arguably makes a real performance difference sometimes, like Intel has with Hyperthreading. They probably fear, rightly, that cheaper desktop dual-core CPUs will cannibalize Opteron sales.
So don’t expect anything too appealing in dual-core from AMD this year.
Dual-core will only hit its stride once 65nm process technology reduces core sizes to something approximating the usual single core size. This is as true for AMD as for Intel.
Intel ought to start making a few 65nm chips for market by around the end of the year, with serious rampup the first half of 2006.
For AMD, building a new fab is much bigger deal than it is for Intel, so progress on the 65nm Fab 36 is relatively more important.
During the conference call, the AMD execs said everything was going fine, but provided a clue by what they mean by “fine.”
“Fine” apparently means, “We’ll be able to get some processors out by the end of the first half, 2006.” Extrapolating from that, considerable production probably won’t come until towards the end of 2006, and crossover into major production won’t probably happen until 2007.
These timeframes are fairly normal; the problem is Intel is ahead, and will probably have its usual six-nine month lead over AMD at 65nm. From a performance standpoint, this will be much less significant than in the past, primarily because AMD has SOI and Intel doesn’t. From a production cost/capacity perspective, though, this is much more significant than it would have been in the past. 65nm will let Intel make lots of dual-cores at a reasonable price without killing fab capacity, which is something that can’t be done by anybody at 90nm. Intel will be able to push dual-cores as the thing to have heavily in 2006, and if it catches on, they’ll have a big edge over AMD for most of the year.
The Real Question Is The Same Old Question
Enough about future problems. AMD’s more immediate challenge is to finally shift production from XP to Hammer and sell millions more Hammers a quarter while trying to retain Intel-like prices.
Can they do it?
It was very clear from the conference call that this is AMD’s intention. AMD’s current gross profit margin is around 55%, and the AMD execs said they wanted to get that above 60%, which is Intel territory (and a goal even they don’t meet sometimes).
Those extra profit points will most likely come from lower costs rather than higher prices, but it certainly is no indicator of planned lower prices.
Of course, you’d expect them to say that. The question is not what they want to do, but rather what they end up having to do.
Up to now, AMD has been able to keep Hammer prices up with little slippage while not making very many. Over the next couple quarters, they’ll come close to tripling the supply.
Can they keep prices more or less where they are, or will demand not meet supply, and AMD have to choose between heavy discounting, or cutting production and losing market share?
Right now, I’d say it’s 50-50 on whether they can keep charging more, and 90-10 that if demand doesn’t meet supply, they’ll choose to cut production first rather than lower prices.
The problem with such predictions is that if the world economies start catching a cold or worse over the next year or so, that can change the whole ballgame and change the odds dramatically.
For right now, maybe the best prediction that can be made is: Bad Times = Good AMD CPU Price.
Of course, if Bad Times = Losing Your Job, a good AMD CPU price isn’t going to help you very much.