HightOwl, "What logic evil INTC was employing when it entered those secret agreements"
I think you are too harsh on Intel. Although I am
not exactly an unbridled supporter of Intel, but I do
tend to think that Intel is a sort of victim here.
Of course, it does not lift much of the responsibility
of Intel management for wrong decision, but the
early Rambus marketing engine was fairy strong.
Since 1990, all their materials were promising
rosed mountains of low cost for fewer pins, cheap
boards, and a leapfrog bandwidth. I have many of
those marketing materials in my files, and they were
really attractive and effectively promoted the
Rambus supremacy ideas. It was really hard to imagine
that the house of seasoned professors and engineers,
assembled to do the only one task of engineering
the superb Ram-bus, could not deliver on promises.
What happened next were "little" difficulties
with implementation of the primary Rambus
ideas of fully-multiplexed bus build in accord with
their early patents. I wonder if the thread realizes
that none of the "Rambus technology" in litigation is
currently in use in industry? That they have to
change the design drastically in order to make
it work and be mass manufacturable? The super-ideal
sharing of control and data lines was backed down
to separate control and data, the memory chips did
not fit into one time domain, and have to be distributed
along several modules creating a new set of difficulties
due to time domain crossing. Only few elements remain.
I am not sure when this new "Direct Rambus" has arrived
and which patents do cover it, but I know of an Intel
software patent on how to initialize a Rambus system.
The patent of Intel, not Rambus. I guess it tells
something.
Finally, the whole idea of 32 impedance
mismatched devices on a single set of wires appears
to be unsound when the signal propagation time
exceeds several clock periods.
It is like trying to run with a long
washtub with water and spill no drop. Of course, it
is possible to model the shape of a simple signal
along the bus wire, maybe even for a few hundred
bit patterns, but some people calculated that the
number of bit pattern combination in reality may
easily exceed the number of elementary particles
in our universe, so you may guess as what kind
of verification coverage this technique has.
In short, the whole multiplexed memory bus supremacy
idea was a big mistake, and many are going to pay for
this. Unfortunately. |
