Hi Gene Parrott; I think that very little will be answered this week, unless Samsung starts RDRAM wafers again. Intel and its buddies bought a lot of material for the Camino roll out, and now they have to move it. They may have to sell it under the cost of production, but they have to move those chips. Until they get rid of it, (at which time the price of RDRAM chips is likely to rise again), there is no way that Intel is going to announce the cancellation of their support for RDRAM. So Rambus is exceedingly likely to be announced. I think that this week ought to be good for RMBS longs, but what do I know.
As far as future memory availability, the thing to look for is an announcement of wafer starts at Samsung. They are the industry leader, as long as they are not starting RDRAM wafers, the box makers have no choice but to assume that this memory technology is dead. That investors do not also make this assumption, an assumption that everyone in the industry is saying, (though not necessarily in public), is something that I do not understand. The writing is very clearly on the wall. No new starts (except for game consoles). Given this situation, where everyone pretty much assumes that RDRAM is never going to be much more than a niche memory, the Intel bundling is easy to understand. You have to look at it from the point of view of a box maker.
Suppose I were a box maker, and suppose I had a bunch of Camino type motherboards (and all the other long lead time parts that would go into such a build), and suppose I wanted to get those parts out the door. What would my situation be, and how would I want to get out of it?
First of all, I would be quite upset with Intel, and not a little worried about having all these parts lying around for a month or two. The better box makers pride themselves on short build times, DELL, in particular, must be just beside themselves with having to store Intel's mistake in their warehouses.
The Camino chips got pulled, and I'm going to need them in order to get rid of my other parts (i.e. motherboards, RSL clock generators, RIMM sockets, etc.) So I definitely need to order them from Intel. The RIMM modules are another case. I really don't know how many I am going to want for each Camino chip. The parts are extremely expensive, but I have reason to suppose that the black market wholesale prices are very low. Otherwise Samsung would still be making wafer starts on them. So if I order RIMM modules too early, I am very likely to end up having way over paid for them. In addition, if problems come up with the Camino, and it gets recalled again, my RIMM modules are going to have cost me more than the whole rest of the motherboard (sans CPU). So I probably don't have a big stockpile of RIMM modules, and will need to add those as well.
The RIMM modules themselves, are typically a late add item. (Since they are expensive, and since they are not soldered down, the manufacturers add them to the board at the last minute, just before final test. This reduces various costs. Most of the reason that high cost parts tend to be socketed is to reduce the amount of time the company has to hold them, not to allow easy replacement.)
If I can get Intel to bundle the RIMM modules with the Camino, then, if the Camino has to be pulled again, I can make Intel eat the RIMM modules as well. Since RDRAM appears to be well on its way to a minor niche market, the last thing I want to have running around my warehouses are a bunch of depreciating RIMM modules, with no Camino to hook them up to. So bundling decreases the risk factors in the technology.
In addition, industry rumors have that the box makers sent their RIMM modules back to the makers when Camino was cancelled. Assuming this is true, then it is clear that the box makers need both RIMM modules and Camino chips. God knows how many they need to keep in stock for memory upgrades, but most users don't upgrade, and right now the thing to do is to get rid of those motherboards at the least risk possible.
Looked at this way, the bundling makes sense. The box makers don't want to make a long term commitment to Rambus, (which would involve making a contractual obligation to purchase a certain number of RIMM modules over some particular length of time) so they want just exactly enough memory and Camino chips to get rid of their already sunk parts (mother board) costs.
It is clear to me that this technology is quite dead, until Samsung opens up the tap. At that time, it becomes a horse race again, but RDRAM will be running a crippled race. Right now, RDRAM isn't moving forward. What you see can be explained as the industry getting rid of the remnants of a mistake. And there is no way that the people in industry who need to get rid of those parts are going to tell you this in advance. Who would buy the boards if they knew that they were not going to be supported in the future?
-- Carl |
