However (and there's always a 'however' with new technologies like this), DDR2 does have some drawbacks. The main one is latency. Memory performance can be characterised as a balance between overall bandwidth and latency. This is the same as with hard drives, where you want a high rotational speed so data can be read quickly, but also a fast seek time, so random data can be found without delay. With memory, high sustained throughput is useful for applications that read sequential data, such as in multimedia, but low latency is especially important for applications that read a lot of random data, such as with servers.
Essentially, because DDR2 has half speed memory chips, its latency does not compare well to that of conventional DDR. While DDR-400 has latencies of up to 2-2-2-5 (where lower is better), the JEDEC spec for DDR2-533 is 4-4-4-12. Write latency is also much slower than conventional DDR. As such, in some applications where bandwidth is critical, DDR2 compares quite favourably to DDR, but in others where latecncy is important, DDR falls short.
In fact, with conventional DDR-533 becoming increasingly popular, it’s probably not going to be until the release of DDR2-667 that we can expect to see DDR2 take the overall lead.
Finally, while CPUs are still riding on an 800MHz front side bus, which has a maximum bandwidth of 6.4GB/s, dual channel DDR-400 can service this quite adequately. When 1GHz+ busses become the norm in 2005, though – that’s when memory is going to be struggling to keep up, and it’s only then that DDR2 will be really viable for the long term.
All this, is not an unexpected outcome. New technologies like DDR2, Serial ATA and PCI Express are engineered with the long term future in mind and as such, the first generation after the prototype usually suffers in comparison to its predecessor.
Ultimately, don’t believe the hype from the memory companies. Hold off on DDR2 for now and wait until there is a really compelling reason to upgrade.