Tony, <1 GHz desktop computers, let alone Willamettes, are ridiculously powerful enough anyway, with plain vanilla memory in them.>
A lot of this was dependent on crystal-ball planning. Back in 1997 or so when the decision to use Rambus was made, I'm sure Intel was thinking that applications in the year 1999 and 2000 will be limited by memory bandwidth. Part of that view was based on history, where Intel noticed that applications in 1997 use a lot more memory bandwidth than applications in 1995 (or whenever). Unfortunately, bandwidth requirements didn't turn out the way Intel planned.
As for 1 GHz desktops with "vanilla memory," remember that part of the reason Intel's 1 GHz Pentium III is a little more powerful than AMD's 1 GHz Athlon is Rambus. Most of the performance benchmark comparisons out there pair up the GHz Pentium III with an 820/RDRAM platform. Meanwhile, the Athlon is still stuck with PC133 SDRAM. Because of that, chipset performance is becoming a bigger issue now with Athlon than before. Of course, AMD will remedy that problem with its upcoming DDR chipset, but that won't appear until Q4 of this year. Until then, Rambus is helping the Pentium III keep up performance-wise with Athlon which has no Rambus support. (Price/performance ratio is another thing, however.)
Elmer is right, in my opinion. Intel's mistake isn't necessarily supporting RDRAM in the first place. Intel's mistake was not having an alternative SDRAM platform available. That should be fixed once the 815E chipset is launched this month.
Tenchusatsu |
