There are three flavors of RAMBUS that will be released. 600, 700, and 800. These chips actually run at 300 Mhz, 356 Mhz, and 400 Mhz. They utilize double data rate (transfering data on the rising and falling clock cycles). Latency is the same across the board for these chips, it is only throughput that is increased. On the i840, with its dual channel architecture, both bandwidth and latency are improved.
As for this memory being overkill for today's processors, that is actually pretty wrong. There are more and more applications coming out that require more and more bandwidth. 3D applications are leading this rush, and it has been shown that increasing the memory and cache output on many of today's applications has shown a marked improvement in performance. Word and Excel will not show much improvement, but I would not consider those bandwidth hogs. You are absolutely correct about latency, but latency is more important in instruction calls and requesting data. Once an application is going full steam ahead, the latency problems tend to be decreased due to how calls are made for memory.
When RAMBUS is increased to 32 bits (as in the i840 chipset), it is very beneficial for things such as AGP 4X (which takes up 1GB of bandwidth when fully utilized), 64 bit PCI slots (which each take up 533 MB/sec), dual or quad CPUs (which each taking up 1 GB/sec of bandwidth). With all of these things added together and running high end applications, the 3.2 GB/sec bandwidth and decreased latency of 32 bit RAMBUS really come into effect.
Proof of what I am saying is easily available, you just need to look it up. In Maximum PC's November issue has a nice little article about RAMBUS on pages 24 and 25. It states that while RAMBUS acts like it runs at 800 Mhz, it is actually just DDR 400 Mhz.
Copied from Raging Bull
by Azure
|
