Thread,
After feeling the pain of some of the more responsible posters on the RMBS thread, I thought we could have a more meaningful dialogue of strengths and weaknesses of the RDRAM/PC133/DDR DRAM memory technologies.
Here is what I think about these technologies, please feel free to knock it down.
The fundamental technology behind all of these memory technologies is DRAM. Assuming comparable process, the latency of this beast is roughly constant between all the flavors. All we are doing is a slapping fancy front-ends to improve bandwidth significantly and probably reduce the latency a little because of finer synchronization.
I consider the latency issue to be a second order effect so from my view point the issue boils down to bandwidth. And, when I look at this that way, what the whole debate boils down to is pincount. Each of the technologies need different amount of IO to offer a set amount of bandwidth.
RDRAM can deliver a set amount of bandwidth with fewer pins but substantially increased design complexity and expense. PC133, DDR DRAM take more pins to deliver the same amount of bandwidth at significantly lower amount of complexity and cost. (Since, cost tends to be a function of volume, the difference between costs will reduce and we can get into end-less debate on what it is going to be in year 2001 but I will not address that in this email)
So, the design choice becomes: how much bandwidth do I need for the design in question and what is the budget for the memory sub-system for the target system.
Based on what we know about the PC architecture and applications today, it is not clear that there are any realistic applications that require bandwidth beyond what PC100 can offer for the mainstream desktop. And for the few applications that are likely to push the edge, PC133 appears to have the ability to deliver the required bandwidth. Need more bandwidth than that, DDR is still a more economic choice than RDRAM. Yes RDRAM does offer more bandwidth than these other solutions today but what applications are there that require this more expensive solution?
Am I missing anything significant here? Can someone take the opposite side and make a case for RDRAM on the desktop today?
Once we hear the other side of this saga, we can cut the religion and go to the next step of analyzing what applications can really use RDRAM today and what applications can use it in the future and how we can quantify the market potential.
Chuck |
