Hi richard surckla; Regarding INTC, DELL, and other OEMs waiting until the eleventh hour to do their testing...
The problem with Rambus is that there just isn't enough margin to make a robust system. Designs which are insufficiently robust can have prototypes that run beautifully, but, when put into production, end up as a disaster. The fact that Rambus doesn't have much margin should have been obvious to everyone with the technical know and experience in memory system design. In fact, this danger was known in the industry. A lot of people knew this. We complained about it. Our complaints fell on deaf ears, for a variety of reasons, mostly having to do with stupidity and greed. This is not something that started a couple weeks ago, this has been going on for years. Why do you think Rambus is so late?
But every time anyone posted a careful, reasoned article from an unbiased industry expert, for instance in EE-Times, they got shouted down. Shouted down by whom? By a bunch of people who were either stupid (in that they were unaware of their limitations) or who had financial reasons to pump the hype (or both).
The problem was that the number of people who can look at a technology and tell that it is going to be marginal at best is very, very, very small. They are a tiny voice and they are outshouted by all the know it alls, most of whom can only quote other know it alls, and rest of whom have no experience in the field that they feel knowledgable about. (I'm trying to use polite words here so that I don't end up in SI jail, but my feelings on this are very, very, very strong.)
Getting design advice from the investor relations people has never been a good idea, marketing and sales people also have only a surface understanding of what it is they market or sell. Upper management, even at the well run companies that made these decisions, understands only surface issues as well. All these people tend to be know it alls.
I am sure that the vast majority of Dell's early production Rambus machines are running beautifully, under normal operating conditions. I, myself, would be willing to buy one of those "defective" RDRAM systems. I would be sure to operate it in a stable temperature environment. I might, for instance, let it warm up for five minutes before making it do anything. I would be very careful if I ever touched those RIMMs, to clean them carefully before putting them back. I would keep dust from getting in the machine. I'd love to have one, at the right price, but I'd also like to wait for an Athlon, which I believe will kick rear end.
I'm certain that I could keep one of those RDRAM machines running, but I'm an engineer and a damn good one. I could also keep a 1974 AMC Gremlin running, for that matter, though I choose not to. In fact, my old 386 machine never did run until I let it warm up for a couple minutes, and this never bothered me, I was hot clocking it, of course.
Machines are like living things, in some ways, and if they are treated with great delicacy, they thrive and last longer. But most people are not like me. The costs to Dell of shipping a system that gets returned is tremendous, they can't afford to alienate customers, much less pay for all that shipping &c. They have a reputation to consider, and that is why they have very high standards.
Once a company ships a product with a known defect, one that only shows up in some small percentage of the product, that defect gets blamed for everything bad that happens afterwards. The Intel screw up on their floating point is a great example. The bug probably changed almost no results for anyone, but every moron who couldn't find the on switch to his new computer blamed the resulting black screen on the bug. So Intel had to fix it.
The worst defects are ones that are rare, as they are the hardest to test for, and once found, they are the hardest to fix. Defects which happen every microsecond are duck soup to fix, its the ones that happen once per week that leave engineers and technicians grasping at straws. These kinds of errors are so rare, that you just can't tell management whether you have fixed it or not until a couple weeks have gone by.
When people start looking for problems, they always find plenty of things wrong. Every computer has a number of errors in its design, all they ever really do is get rid of the more common ones. For instance, bus timing might be a tad bit too tight, due to someone not derating an output driving a particularly high load. But there is enough margin in the system that the design error never actually causes a failure in any system that is ever built throughout the product's entire life cycle. But with a design error of that type, if enough parts are at the limits of their guaranteed specifications, the error can come through. These kinds of things are just not possible to eliminate from designs, though God knows we try. This is the bete noir that wakes design engineers up shaking at the time that designs are released to manufacturing.
At least half the art of designing electronics for high volume manufacturing is to be sure that the design has enough margin to allow human error to keep it running. I know that engineers include margins, that is why I hot clock my computers. But if the manufacturer hot clocked every machine he shipped, he would be in a world of hurt.
-- Carl |
