Hi John Walliker; Re those possible driver problems. I guess I agree with the HD engineer on this one.
Driver problems result in lines that are not driven correctly. The longer the line, the worse the problem.
If the tranmission lines were perfect transmission lines, this would not be the case, the lousy edges would just propagate down the wire with no changes. But computers are not built from perfect parts. For every part, there is a spread in parameters. The art of electronic design (for high production as opposed to prototype work) is to create a recipe for connecting things (i.e. motherboard, chips, resistors, etc.) in such a way that it has to work.
Worst case design means that the design has to produce a working board when it is built from any parts that meet the specifications for that part. All parts have specification ranges, that is, no specification on any part is every exact (unless it is an integer, like the number of bits), instead all specifications have a range. For instance, a propagation delay may be specified to be somewhere between 2.0 and 3.5ns.
So to prove a worst case design means that you have to verify that the design will still function despite its being implemented with part A being fast, but part B slow, while part C is medium, for instance. Since each part has a fairly large number of specifications, checking a design for worst case is a fairly involved activity.
Getting a prototype to run does almost nothing to prove that a design works, though if the prototype doesn't work, the design is definitely in trouble. All a functioning prototype proves is that a particular set of parts, (for instance this capacitor, marked as 100pF, but actually approximately 97.23212345pF) works when running a particular program, at a particular temperature, given a very particular electromagnetical noise environment, given etc., etc., etc. There are a way more than just infinite (in the mathematical sense of differing orders of infinity,) number of possible conditions that a design couuld be assembled under, a prototype just tests one or a few of these.
The Rambus design relies on a transmission line, so engineers have to figure out the range of possible variances in all the components that make up that transmission line. That means the range of possible capacitances on input pins of RDRAMs, the range of possible inductances of traces on RIMM modules, the range of resistor values, etc. The list would be quite lengthy, and this is why the resulting tranmission lines will not be a perfect ones. This is not something unusual about rambus, it is the normal condition of engineering design in a world that is most notable for its deviation from perfection. (As opposed to God who, I understand, works with perfect 1 ohm resistors. If He needs a 1M ohm resistor, he just connects up one million of his perfect 1 ohm resistors, which, having no inductance or capacitance, can be joined by Him arbitrarily. He can also produce resistors with resistances of exactly Pi ohms, but I do not know if he bothers to do this.)
But on this frail and mortal Earth, engineers traditionally add a little extra margin on all their parts requirements in order to account for things that they forgot to model. The hubris of the Rambus design teams is to forget to add that margin. These systems were being built with insufficient margin, and consequently they are buggy.
What the Rambus guys did is to try to cancel long chains of propagation delays. Cancelling prop delays is a ancient and revered technique for engineers to approach the Godly perfection of perfect parts, but engineers have to remember that they are mortal, and so they have to give themselves, and by extension, their parts, a little room for error.
It's all about sin, really. The rambus hubris was to assume that Man is without sin.
-- Carl
Some of this may have been confusing. If so, just look at the stock price for an explanation. |
