Hi all; I'm feeling a bit guilty for letting the more paranoid section of society believe that I'm being paid to bash Rambus. As far as I know, no one is paying anyone to bash anything anywhere, certainly I've never had anyone ever even offer to pay me or anyone else to do that. Sorry for deliberately making fun of that stuff, the truth is that a lot of you are truly stupid and will believe pretty much anything, I shouldn't make such sport of you. My favorite was the guy who used the word "payed". (What has happened to education in this country?)
On the other hand, people get stuck in jail for being paid to pump stocks with some regularity. The typical case is a bulletin board newsletter investment advisor who fails to make public the fact that he is being paid in stock by the company that he is recommending.
Why doesn't someone find a basher complaint in the following list? Every action I looke at was perpetrated by people long the stock, but there are lot of them, who knows:
securities.stanford.edu
Someone please give me a link to someone who was proved in a court of law to be paid to bash a stock. There might be someone out there who shorted a stock and then bashed it, of that I have no doubt, but not someone getting paid to bash.
I think that the technological inadequacies of RDRAM have been as fully demonstrated on this thread as any reasonable person could desire. The next three months will see the unfolding of DDR systems for the masses, the bench marks will speak for themselves. I am quite sure that DDR will be here to stay, and its technological superiority will prevent Rambus and Intel from getting their duopoly. The story is over.
I should repeat that logic. The demand for different types of memory chips is driven by design engineers. The primary motivating factor of design engineers is cost reduction, provided that the design solution does its job. But at any given level of performance, the primary job of the design engineer is to design a working product that will be the cheapest to construct, over the lifetime of the product. This is in distinction to the task of more average engineers, not designing for mass production, which is to quickly design something that works, pretty much regardless of how much it costs to build.
The big fear of design engineers is that they will design into a product a memory type that will become an orphan during the lifetime of that product. An example of an orphan memory type is the SLDRAM. The guys who designed that into systems without recourse to SDRAM replacement have to be hurting pretty bad right now, I would think that the parts are hard to find and pretty much single sourced. So why do I think that DDR is any different from SLDRAM? After all, they were both vaporware at one time.
Different products have different product lifetimes. The graphics gaming accelerator cards (i.e. Nvidia's GeForce) guys have the shortest product lifetimes, only around 6 or 9 months. As an example, the original GeForce is already obsolete and replaced by the GeForce2, only six months after its introduction. These guys have to be sure that their memory chips are going to be available for only that short period of time, so they are not a leading indicator for whether a technology is going to be around a long time. But the fact that the X-Box and Dolphin are both DDR design wins is a pretty good indication that the stuff is going to be around for 3 years or so, at least for game consoles.
The way that design engineers get a feeling for whether a memory type is going to be around is by talking to the memory makers. What you want to see is several different suppliers all saying that they are going to have compatible parts, available to your company, over the time frame that you need them, and at the right price. That is why the memory industry typically arranges for quick second sources of new memory types, rather than trying to establish a monopoly. If enough memory makers say they will support the chip, then the designers tend to believe them, and will go ahead and design the specialized memory type in.
At this point, production of the x32 DDR chips are assured for the next 18 months. After that, they will go to DDR-2, with data rates of 600MHz. This should keep the graphics industry happy for the next 3 or 4 years or so.
The other major market that has gone completely to DDR (to the more or less total exclusion of RDRAM) is the server market. Completely unlike the graphics market, the server market requires products with very very long lifetimes, on the order of 3 years or more. For this reason, it is very significant that the server guys went with DDR. If any memory designers in the electronics industry were to be convinced that DDR would orphan, the server guys would have been them. They have the most to lose, yet they all went with DDR, even Dell.
The memory makers convinced the server guys to believe that DDR was here to stay by demonstrating that they could make DDR on the same fabrication lines and using the same masks as SDRAM. Since everybody knows that SDRAM will continue to be produced for the next 4 years, that means that DDR can also be built, providing there is a market for it. The server designer guys heard the same story from all the memory makers and they believed it, so they started designing DDR into their new systems.
Incidentally, this is in stark contrast to the SLDRAM situation, where only a few (if any?) engineers designed the product into their systems. Those engineers, by failing to follow the mass of other engineers, ended up out in the cold.
Now that the server guys went 100% for DDR, the rest of the industry knows that DDR memories are going to be available for the next five years, and not just for servers. That long term availability is mentioned in the Micron advertising literature.
But it is also at least as clear that RDRAM will be available for, at the very least, the next two years, why did so many memory designers choose DDR? The memory makers have done everything but promise that RDRAM will not get as cheap as DDR / SDRAM. So if it all comes down to price, designers will choose DDR, which is any case easier to use, in many ways, than RDRAM. In addition, there are plenty of indications that the supply of RDRAMs will be constrained for some time into the future. On the other hand, DDR are built with only a minor late change to SDRAM wafers, so they are likely to be available with a much smaller lead time than RDRAM. Designers know this, so they're only slightly more worried about getting supplies of DDR than they are about getting supplies of SDRAM. But the basic fact is that mass users of memory don't generally buy parts on the spot market. Instead, they buy it direct from the memory makers, making contracts in advance. This allows the memory makers the ability to prevent themselves from ending up in a glut of unwanted parts. Incidentally, the glut of RDRAM chips that was a consequence of last year's Camino fiasco decreased the inclination of memory makers to quote that part at cheap prices to box makers. Instead, the memory makers learned to quote in a risk premium. DDR runs don't need the risk premium because if the buyer doesn't want them, they can be turned into SDRAM at the last metal layer. It is my belief that the low risk premium is why the memory makers are quoting DDR to the market at a very small premium to SDRAM.
I've been pretty thorough with regard to the technological deficiencies of RDRAM over the past two months. For this reason, I am going to largely leave this thread to its own devices for a while. Look for me again sometime this summer, when the cr@p finally hits the fan, though, I do look forward to again saying "I told you so."
-- Carl |
