I was just over @ acesharware and they posted some pretty interesting numbers on single channel DRDRAM. The tests were done on a p4 system, running sandra 2001. Basically with only a single channel DRDRAM the p4's performance numbers were halved.
Fpu 1431 700
cpu 1381 650
The dual channel has a bandwith limit of 3.2GB/S vers. the single channels 1.6 GB/S.
My point to all this info is :
1) p4 archtecture was designed for and really needs DRDRAM to perform as designed
2) DDR pc2100 has .5GB/s advantage over single channel RDRAM
3) Rambus/Intels failure to move DRDRAM into mainstream means 2001 intel must start using DDR RAM for p4.
4) performance of p4 may be even more poor than it is now because it must use a memory technology that has bandwidth similiar to single channel DRDRAM.
So as DRDRAM becomes a faint memory in 2H 2001 for mainstream computers. We might see that intels adoption of DDR will handicap the P4.(even with new sse 2 optimized software)
Do remember that bandwith is not everything nowadays. There is a big problem concerning RAMBUS that is called latency.
There is a very good article released on [H]ardOCP. Everyone should read this before they begin talking about bandwith!!!
<A HREF="http://www.hardocp.com" target="_new">http://www.hardocp.com</A>
Read the article on bandwidth as soon as it came out(unlike most people, I Kill bias by reading many articles) and I agree with the author that latency is the real problem with gigahertz processors of today and tommorrow.
DDR is in it's early phases and as it evolves and we get in to DDR 1.5 /2.0 latency issues will start to get addressed better(bandwidth will also be increased).
DRDRAM whether pc600 or pc1000 has latency issues that will probably not get addressed for several reasons:
1)DRDRAM is NOT going to be in mainstream computers any time soon or ever(I think most will agree with that)
2)due to reason #1, stocks of pc600 DRDRAM are probably still plentiful(that is why box makers are curently using them)
3)Once supplies are depleted ddr will be the standard for most if not all mainstream computers(many large memory companys are not even making DRDRAM anymore)
And Finally 4) why address latency issues with DRDRAM when there will be no financial gain for Rambus due to the above reasons(NO market->just park it(HEHE!))
So your right, bandwidth is not the major problem for much of todays computing, it is latency.
My question to intel is why did they design an architecture(i.e. P4) that so heavily relys on bandwidth.
I guess they gambled and designed a processor to take advantage of a memory technology they thought would be in mainstream pc's.
Well it's time for intel execs to go to gamblers anonymous.
"How in the heck are people testing a P4 system using DDR SDRAM when Intel doesn't even support it in any way yet? I'd say that any such results are completely bogus, if not just for that reason alone."
If you reread the original post they are testing using single channel RDRAM.
The article is of interest because it shows how dependant the P4 is on memory bandwidth in some tests. More worryingly the test was carried out on a chipset which will be released meaning that some computers may have this memory configuration. I think even you will agree this configuration would have an extremely bad price/performance ratio.