Hi all; Re RDRAM being dead, dead, dead...
At this year's Platform 2000, Transmeta's head guy gave a talk on memory in mobile platforms. He prefaced his talk with a discussion of why RDRAM is dead as a technology. I suppose that this was so that he didn't have to bother with it in the rest of the talk. So if you want to hear a (disembodied) voice saying pretty much the same things I've been saying re RDRAM and next year's technology, here is your chance:
Memory Solutions for Mobile Platforms
Bill Gervasi
Technology Analyst, Transmeta Corp.
Chairman, JEDEC Memory Timing
03:30 Rambus PostMortem
Postmortem on a memory technology that is in the process of dying quite noisely.
06:00 Platform 99 Projections
Here's what I said last September would be the roadmap moving forward. (Projections as of 07/99)
What I anticipated was that segement 1 and segment 2 of the PC marketplace would be split between Rambus and DDR moving forward. I anticipated that Intel would be successful in capturing this segment here, and then about six months later would be moving that success into a lower priced segment of the PC market space. Other factors were a transition from a single data rate 167MHz for graphics, and making a quick transition to a x16 DDR SDRAM, and then eventually, about six months to a year later, the graphics guys would cost reduce, going to the more price sensitive market with a x32 DDR SDRAM. That was my vision in September of 1999.
I've updated these projections. Certainly the biggest factor is that I no longer see Rambus moving beyond the high end of the PC Segment 2 space. All the noise on the rail is that it is essentially dead, and that resources are already being pulled from this. What we said we'd see was some PC100, maybe PC133 happening in Segment 2, but it looks like a pretty quick transition to DDR instead of Rambus is more likely.
11:37 Market Factors
Rambus appears to be dead...
16:00 Rambus Mistakes
*Performance < PC100
*Systems too costly
*Answered 1994 pin count problem, but
* 100PQFP gave way to 400 BGA
* Required grounds eliminated advantage
*RAC does not shrink
* Die penalties increase with shrinks
Memory guys don't know how to shrink big analog things.
*Power centered in one chip at a time
19:00 DDR Beat Out Rambus
DIMM cost -$10
PCB cost -15%
Latency -27%
Peak BW +33%
Power -40%
silicontech.com
As usual, it is likely that I mistyped or misheard some of the above dialog. So go to the original source and hear it yourself.
-- Carl |
