Hi Ali Chen; I guess I should clear up a few of jim kelley's putting words into my mouth with regard to transmission lines &c.
I don't have anything against transmission lines, I use them all the time. Even transmission lines on PCBs aren't that much of a problem. Where RDRAM gets into trouble is with all the stuff attached to the transmission line. I don't know what kind of transmission lines DDR-II is going to use, but I am certain that they are going to be a whole lot more robust than the RDRAM garbage.
In addition to 32 RDRAM chips, there are a driver, a termination resistor, and up to six (was originally supposed to be eight) socket connections. This involves something like up to 39 solder joints, and up to 6 board/connector connections.
Even then, they could get them to work, but in addition, they wanted them to be compatible between multiple manufacturers of the memory chips, up to 16 different numbers of RDRAM chips on each and therefore RIMM type, several different generations (i.e. 64Mb, 128Mb, etc.) of memory chips, and multiple different memory controllers.
Even that could be done, but in addition to all the above, they wanted to send data down the whole shebang at 800MHz, with 33 lines in parallel, and bidirectional, and using a standard CMOS process. No way.
As an example of a reasonable use of transmission lines, take AMD's Lightening Data Transport. It has only two chips on a wire, it single directional, no connectors, and are at a much lower frequency. (And is therefore easily and cheaply manufacturable.)
Quite frankly, RDRAM was an idea thought up by a bunch of impractical over-educated know nothings. We've all seen these kinds of people in action at technical companies. They show up with a big pile of credentials but no real understanding of the practical problems of the real world, and unless management is capable of recognizing engineering, they destroy whole projects and even companies.
The technical management at Intel had drifted to a sufficiently low level that they were unable to recognize Rambus for what it was, and viola, they bought it. Then the rest of the industry had to make gambles depending on whether or not they thought Intel had the ability to force RDRAM to be the next standard. Voila, a couple of extra design wins.
If RDRAM had gotten its act together 3 years before Camino shipped, they could have gone with 566MHz (equivalent to PC133) and maybe have survived. But they bit off more than they could chew (maybe because of the reduced power standard for RDRAM was incompatible with the speeds required on the larger gate lengths?), and the technology had to be delayed to 1999.
And since DDR and PC133 were on the design blocks, they couldn't simply get by with PC566/600. So they had to go with PC800 and they were unable to deliver. Now it is too late, the game is over. By the end of the year, it will be clear to all that DDR will be the new memory standard.
-- Carl
P.S. I read somewhere that Sun was in development on a DDR chipset. Have you heard anything about it? I would think that they would be working pretty hard on a replacement memory interface for that MAJC, either PC133 or DDR266, but they could also be working on something for one of the Sparcs. |
