jhg,
<<33 pins -- why is it critical>>
Drats...more homework! LOL OK, I'll try.
There are several ways to increase memory bandwidth. One way is to increase the memory clock speed. Another is to create
additional channels. There are other ways that don't concern us.
1. Real estate is at a premium for workstation and desktop PC
motherboards. Therefore, adding additional SDRAM channels,
168 traces wide, isn't feasible. On server boards this isn't
a problem. One could put 4 Rambus channels and use only
132 traces. 4 Rambus channels give a bandwidth of
6.4 GB/s/132 pins. Rambus provides "headroom" for the
future. Notice that we haven't used as many pins as a single
SDRAM channel which transfers at a peak bandwidth of
2 GB/s/168 pins. This doesn't even take into account the
efficiencies of the technologies (Rambus 95%, DDR 60%).
2. One might ask, "Why not just increase the clock speed for
DDR memory?" Well, as one increases the clock rate, the
memory channels begin exhibiting transmission line effects
(reflections, distributed impedance characteristics, etc).
Therefore, this is the fantastic advantage Rambus has. RDRAM
is designed from the ground up taking transmission line
effects into account. This is a fundamental difference.
Hope this helps you to understand why I am so bullish on Rambus.
One other little tidbit I would like to pass along...I am reading
The Gorilla Game, Picking Winners In High Technology, in which
the authors state, "in 1998 a company called Rambus took on a
gorilla position in the emerging market for next generation
memories." Page 135 of the revised edition if anyone is interested.
It's kinda nice to have my views confirmed in this manner.
Regards,
Glenda |
