To: unclewest who wrote (25656 ) 7/25/1999 1:12:00 PM From: Matrix_Man Read Replies (3) | Respond to
Uncle - Rambus' solution, although touted as a general solution to the DRAM data rate problem, really only addresses a subset of this problem. Rambus really only works for small DRAM sizes typical in desktop or embedded systems. When moving to server sized memory arrays, Rambus provides no better pin data rates overall than soon to be available DDR-SDRAM. Consequently, it is not really the preferred choice for implementing server memory systems. Except for the fact that Intel chipsets will only support Rambus in the future, most PC servers would probably not use Rambus. In addition, although technically, Rambus makes sense for desktop and embedded systems such as graphics due to its higher level of integration, it has a high cost due to traditional SDRAMs due to the overhead of its link drivers. Thus, for low end systems where is makes the most technical sense, it makes the least business sense. Thus, Rambus success really lies in Intel's commitment to supporting it as the primary choice for future PC memory systems. As volume goes up costs should go down, but will never be as low as standard SDRAMs due to the licensing costs that Rambus charges. However, Rambus future is still quite good because, as has been mentioned throughout this paper, those companies that are paying the price of overcoming the high cost and power consumption of high speed signalling now will be better off in the future, since it is inevitable that anybody that wants to be in this business will have to go through this same process.
