Thanks for the article, Jim. According to Toshiba, DDR-266 has a effective sustained bandwidth of 897 MB/sec, while RDRAM has a effective sustained bandwidth of 1190 MB/sec. So Toshiba thinks RDRAM has better real-world bandwidth, even though DDR boasts of a higher theoretical bandwidth.
To be sure, sustained bandwidth is the result of a wide, wide variety of factors, including traffic pattern, application and benchmark used, and chipset design. Taking chipset design as an example, there are ways to work around DDR's inefficient use of bandwidth. And likewise, there are also ways to work around RDRAM's longer "core" latency. As for choice of application, surely any benchmark which only access memory once in a blue moon will favor DDR over RDRAM. (But then, that same benchmark will also show no difference between DDR and cheaper PC133.) But benchmarks which create more complicated traffic patterns will probably favor RDRAM over DDR because of RDRAM's more efficient use of bandwidth.
All in all, I find Toshiba's white paper to be very informative. I agree with every conclusion put forth by the white paper, that DDR will become the memory of choice for certain applications, and RDRAM will be for others. (FCRAM is also tossed into the conclusions, but I don't know much about it to form any opinions).
Tenchusatsu |
