Hi All; Re RDRAM being dead, dead, dead. Funny thing, the stock goes up a few percent, and the bulls start thinking they're technological geniuses. Nope. You guys still don't know jack about memory design, and never will.
No more conferences until July, I think, so it's probably a good time to put out a more complete description of why memory designers are no longer seriously considering RDRAM for new designs.
WHY RDRAM IS DEAD, DEAD, DEAD
All technologies die eventually. From the point of view of designers, a technology is "dead" when designers quit choosing it for new designs. This is something that happens about a year before the last design wins for the technology are publicly announced. The reason for this delay is that designers typically choose the memory type for a design early in the design process, and then cannot easily change it later. Companies do not generally announce their designs when they start them, but instead when they are reasonably close to completing them, or if they are a big player in their market, when they are close to shipping product to customers. The big players don't announce products early because they lose customers for their current products. Small players, on the other hand, will frequently announce products way before they ship. In either case, there is an appreciable lag between the design community dumping a technology and the rest of the world finally realizing that it is dead. Designers know these things first, but the rest of the world can figure it out at about the same time by carefully reading the same stuff that the designers read.
The main task of a designer is to produce a design that works. That is, the computer, for instance, has to compute. The second task of a designer is to produce a design that makes an inexpensive product. Companies are in the business of making money, and it is this second factor that influenced designers away from RDRAM. Simply put, it became obvious that DDR would be cheaper (particularly as a total system cost) than RDRAM, and that it would always be that way. Since RDRAM also implies much higher system costs (i.e. for things other than the memory chips themselves), this made the design decision obvious. Don't choose RDRAM. Instead, designers are choosing between various versions of SDRAM and DDR. RDRAM isn't even on the list.
The current price of RDRAM is way too high. I've said before that I expect it to get down to as low as 50% over SDRAM, but that is still way too high. But what memory costs today is not that big of a concern to a memory designer who is deciding today what memory to design in to a product to be built in the future. Chipset designers make memory choices as much as 2 years or more ahead of any production memory purchases. The reason for the long delay is that it takes that long to get the designs finished and into production. Graphics and embedded systems designers, may have a shorter design time. Motherboard designs are usually faster yet, typically started well within a year of actual production. Memory prices can fluctuate quite a bit over the space of a year:
members.home.com
Fortunately, designers don't have to predict the absolute prices of memory in advance. They just have to guess the relative prices. That is, they have to choose the parts that are going to be cheapest. This is a lot easier than it sounds, I will try to explain how this is done.
Rules for cost effective memory selection.
(1) Avoid ending up requiring a memory that is only available from less than four suppliers. The dearth of suppliers means that prices are likely to remain high. In other words, go with the commodity parts. RDRAM does not yet have 4 major manufacturers yet, and the price is sky high. More should come on line this year, so this is not a real concern for the designer making his choice today.
(2) Choose the cheaper package. Generally, this means use the older package until the technology no longer fits into it. An example of this is provided by the packages that the x16 and x32 SDRAMs come in. The x16 package is older, (and smaller), and is going to be cheaper. So in the absence of other influences (and there could likely be other influences), choose the x16 package. On the other hand, the x4, and x8 SDRAMs all come in the same package as the x16s, so choose whichever you like, they will cost about the same. Note that RDRAM has a very expensive package. Someday this will come down in price and will be a good choice for DDR-2 or DDR-1.5, but not for the next few years.
(3) Look up what the memory makers say about the future relative prices of the various choices. Surprisingly, they are pretty good at this, and will tell you. If you are a very big user, they will be happy to sign a deal with you where you guarantee to purchase a given amount over a period of time, and they agree to sell it to you at what seems like a pretty good price at the time. This is probably how Dell got its relatively cheap RDRAM. The memory makers are now talking about DDR being a few percent above SDRAM pricing by 3Q00. When they say this, they are talking about the pricing they give big customers who put in that kind of big orders in advance. When Intel tried to get them to sell RDRAM at 30% over SDRAM a few weeks ago, they refused. Given how much SDRAM has gone up since then, one guesses that there was more than just one reason for their refusing to deal with Intel. In any case, the memory makers have repeatedly stated that RDRAM is expensive to manufacture. This is about as big a clue as they can possibly give a designer.
(4) Make sure your design is compatible with as many memory types as possible. As an example, different manufacturers will frequently require different power-up initialization for their memory. In that case, create an initialization circuit that will satisfy all the chips requirements. (This is a lot easier to do than it sounds.)
(5) It is okay to design in a rare memory if it is compatible with a different memory (with less performance) that is available from more suppliers. VIA provides two examples of this principle. (a) Their motherboards that support NEC's VCM also can use SDRAM (and that is what the vast majority of them ship with). (b) Their new DDR chips will allow motherboards that can be stuffed for either SDRAM or DDR DIMM support. A special case of this rule is that you should try to make your design compatible with as many as possible of the different speed grades. That way procurement and marketing can choose which speed grade (and possibly system performance) to ship according to the market prices at the time. Another example of this rule in action is FCM. Toshiba is making it compatible to DDR and SDRAM in order to hope that it sneaks into a few design because of this rule.
Of the above rules, RDRAM passes rule #1, that is, there will soon be 4 or more suppliers on line. It fails rule #2, as it is in an expensive package. It is possible that Micron will bend this a little, but you would have to have the inexpensive packaging available from four memory makers before you could be sure that it would result in cheap parts. (Until then, it just makes Micron happy, as they slightly undercut the competition and make nice profits.) In addition, RDRAM fails rule #3, the memory manufacturers are bad mouthing it big time. Rule #4 doesn't apply to it. And since, RDRAM is highly incompatible with prior generations of memory, it isn't possible to squeeze it in under rule #5.
-- Carl
P.S. No sweat here. Also, I am hardly the only person who knows that RDRAM is now dead, and is willing to say it out loud. For example, listen to Bill Gervasi, Transmeta Technology Analyst, mention that RDRAM is dead (and explain why) in the first part of his presentation, and then ignore it for the remainder, at the recent Platform 2000 conference:
(audio): silicontech.com |
