Sure, I'll be happy to state my opinions on conversion to RDRAM, but they are just opinions. First let's establish my underlying premises:
1. In a PC environment, RDRAM provides no measurable performance advantages over PC133 SDRAM when combined with a L1/L2 cache except for highly unusual applications. Most programs do significant amounts of loops and the cache hit rates run about 90% or better, so for those programs any advantage would be reduced by 90%. I knew a guy once who had a special program with a single 16mb loop that was executed over and over, and for his application faster memory bandwidth would provide a linear improvement in overall run times. For most applications the difference will be in the 1-2% range, if at all.
2. While it is true that with RDRAM there are savings that come from pin reductions and from an elimination of L2 cache, the overall cost of a system with RDRAM main memory will be significantly higher than the overall cost of an SDRAM system.
3. Memory makers (including MU) could produce RDRAM in quantity if the demand exists, though the price will need to be significantly higher than the price of SDRAM in order to motivate them to do so.
4. Intel is committed to RDRAM.
5. Consumers are stupid.
Based on these premises my opinion is that the RDRAM conversion will be interesting. Companies that do a fair amount of R&D and that are known to be independent thinkers, such as CPQ and IBM, are no doubt well aware of the lack of performance gains from RDRAM and the excess costs, and they will be somewhat slow to offer much in the RDRAM area, except perhaps in servers and workstations. On the other hand, companies strictly follow the Intel line such as Dell will probably do what Intel tells them to do and offer RDRAM at once broadly across the product line.
If I am correct, RDRAM will then become a marketing issue, to be decided by the consumer, with assistance from magazines. I don't put much faith in consumers, and I suspect that 500MHz RDRAM will sound more impressive to them that PC133 SDRAM, and since 500 is greater than 133 they will assume that the system with RDRAM is 4 times as fast. If consumers are as stupid as I think they are, RDRAM has a fighting chance.
As for your other questions, yes I presume that MU still hates Rambus, but that they will make whatever chips are the most profitable. Any production of RDRAM automatically reduces SDRAM output, but presumably the price will be high enough that revenue would rise. I have no comments on your other questions.
Carl
|
