Tenchusatsu heeft op de Silicon Investor forums een errug handige vergelijking met de voor- en nadelen van Direct Rambus DRAM en DDR-SDRAM in mekaar gemikt:
+ Low number of pins: Excellent for chipsets, less expensive north bridge design, also excellent for having two or four RDRAM controllers in parallel
+ Finer granularity: As memory density increases, fewer chips can be used.
+ Smaller packaging: (Or so I'm told; I can't be sure.)
+ Efficient in bandwidth utilization.
+ Great for systems which don't need high capacities (gigabytes) of memory.
- High prices: Hopefully, this should come down as RIMM production ramps up.
- Huge transition hump from SDRAM to RDRAM production.
- Lower max bandwidth compared to DDR, but efficiency makes up for it.
- Slightly longer latency: For desktops, this can have a noticeable impact on performance, though I believe the chipset makes a much bigger difference on latency than RDRAM itself.
+ Higher capacity limit per channel: RDRAM only allows for 32 chips per channel, but with double-sided, double-stacked DIMMs, SDRAM allows for 128 chips per channel. This leads to 4x the max capacity.
+ New density technologies (256 Mbit coming up, 512 Mbit after that) will probably show up on SDRAM before RDRAM.
+ Easier to implement ECC and chipkill features than on RDRAM. That's critical for high RAS in servers.
+ Higher max bandwidth than RDRAM.
+ Slightly lower latency than RDRAM.
+ Likely cheaper to manufacture.
- Pin count requirements are significant. I think it's 2 to 3 times that of RDRAM.
- Because of pin count, it's harder to group two or more DDR channels in parallel.
- Possible electrical concerns: Despite the difficulties implementing RDRAM, implementing DDR isn't going to be easy.
- Coarser granularity than RDRAM. You can have fewer than eight chips on a DIMM, but the packaging will be larger and more expensive per chip.
- Less efficient at utilizing bandwidth, which negates DDR's max bandwidth advantage over RDRAM.