Okay, originally this was just going to expand on a few of Crashman's points. However, I've decided instead to just cover the memory basics in general since I'm not convinced that it is common knowledge with the very partial views being represented here.
<b>1)</b> The P4 was not 'designed' for RDRAM. The P4 was designed for high bandwidth. RDRAM just happened to be the only memory technology able to provide that bandwidth at the time it was developed. However, there is absolutely nothing preventing another memory technology from becoming better than RDRAM for a P4.
<b>2)</b> Each upgrade in memory (IE from DDR333 to DDR400, from PC800 to PC1066) offers more than <i>just</i> a bandwidth increase. It also offers a latency reduction. So while the <i>bandwidth</i> of DDR333 is wasted on an AXP with a 133MHz FSB, the <i>lower latency</i> is <b>not</b> wasted.
This is why the AXP does see minor improvements to performance <i>even when the memory bandwidth is being wasted</i>.
<b>3)</b> DDR SDRAM <i>does</i> have a lower latency than RDRAM and thus performs better at <i>equal</i> bandwidths. However, RDRAM's advantage is in it's ability to scale to higher bandwidths very easily. There is no mainstream DDR solution to match PC1066 yet, and should Intel or AMD implement an 800MHz or higher FSB soon (without decreasing the effective bandwidth per cycle anyway) DDR's continued use is a complete gamble, but RDRAM is much more likely to be able to meet those needs.
So long as the Athlon <i>isn't</i> driven by bandwidth, memory with better latency will always be better for the Athlon. (Currently DDR SDRAM is this memory.)
So long as the P4 <i>is</i> driven by bandwidth, memory with more bandwidth will always be better for the P4. (Currently RDRAM is this memory.)
And in the case where multiple memory technologies fulfill the complete bandwidth needs of a CPU, the lower latency memory will win in performance. (DDR400 vs. PC800)
<b>4)</b> Dual-Channel DDR SDRAM is pushing the wire count pretty far. It's causing considerable consternation to motherboard manufacturers because of the sheer amount of signal noise and crosstalk that they have to fight against.
RDRAM has its own crosses to bear, however insane wire counts are not one of them. RDRAM, being a <i>serial</i> technology, has a vastly lower wire count.
Serial technology is just plain easier to implement, hence Serial ATA replacing the clunky IDE cables.
And thus, the future of memory depends entirely upon how much memory bandwidth is pushed, and how well DDR can overcome signal noise and motherboard complexity problems.
If dual-channel DDR's bandwidth can be increased (such as QDR and ODR) <i>without</i> a significant increase in the number of wires and without a significant increase in the signal noise, then DDR will survive if there is another huge push for bandwidth.
If not, RDRAM will be <i>the</i> memory technology to survive another huge push for bandwidth.
Of course, should there be no more huge advances in CPU bandwidth, then DDR will eventually crush RDRAM because of it's superior lower latency.
<b>5)</b> Conclusion? There simply is none. The memory technology of now depends entirely upon your CPU since Intel and AMD have <i>very</i> different ideals of CPU bandwidth. And worse, the future has even more variables to affect the outcome of which memory is needed where. So just sit back and enjoy the ride. However, my guess is that the memory technology of the future isn't even going to be a derivative of any currently in use.
<pre><A HREF="http://www.nuklearpower.com/comic/186.htm" target="_new"><font color=red>It's all relative...</font color=red></A></pre><p>