What is the difference between RDRAM/Rambus memory and DDR memory?
Rambus / RDRAM runs at data rates up to 800 MHz (the actual memory clock is 400MHz, but data is transferred on both clock edges similar to DDR memory), but unlike SDRAM and DDR SDRAM, which have 64-bit data paths, it only has a 16-bit data path; so its theoretical maximum bandwidth of 1.6GB/sec is the same as the theoretical maximum bandwidth of PC1600 DDR SDRAM (200MHz x 64-bits or 8 bytes per cycle = 1.6GB/sec). Latency for RDRAM is also higher for the first few accesses--the protocol-based approach used by RDRAM requires a certain amount of overhead that actually increases the amount of time for the first memory access.
Despite its higher initial access latency, RDRAM bandwidth utilization is more efficient than DDR SDRAM. One key to this efficiency is the number of sense amplifiers in RDRAM--RDRAM can keep many more pages charged at one time than DDR SDRAM. The high-end Intel i840 RDRAM controller can keep 32 pages open simultaneously, and RDRAM technology has the potential to have many more open pages. Assuming a low number of column and row switches, Direct Rambus DRAM has better overall bandwidth compared to DDR.