RE: <I am curious however, about your change from being pro-Rambus to being anti-Rambus over the past couple of months, especially given the fact that you had changed to an anti-Rambus stance a few weeks before the announced 820 chipset delay. Would you be kind enough to share with this board the rationale for your change of heart?>
[I answered this yesterday but it didn't post due to SI or BrowseMaster having problems. But I've finally recovered it and here it is.]
I wish that I could explain it in a short, concise way that everyone could understand. I don't have much time right now but I'll take a crack at it:
12 months ago I was of this opinion: anything that saves pins would be a great idea, anything that improves memory performance would be a great idea, the dram companies are brain dead and will never come up with anything good on their
own so someone has to tell them what to do. I was a typical, arrogant
microprocessor designer who had not really examined any of this carefully but thought I knew everything about it anyway.
About 12 months ago I happened to talk to several dram people and listened to them long enough to have a seed of doubt planted in my brain. Things like: minimizing cost per bit is the charter for the dram industry and pays off more than anything else, latency in more important than bandwidth for most of the market, granularity is not
close to hitting the PC market, mux'ing data down to 16b just to demux it back to 64b doesn't make much sense. Also the fact that Rambus requires them to undergo a huge upheavel with new designs, new testers, and a significant die size penalty.
But I didn't think too much about it until about 6 months ago when I learned that Intel was going into the Coppermine/Rambus launch with no Sdram option. I was very upset about this as a pro-Intel person. Probably no one on this thread will believe it but I've always been an Intel person, having worked there 8.5 years and still having many, many friends there and often being an Intel investor. I was shocked
that they would risk everything on something that wasn't proven. So I started to follow it more closely and, at some point, started following this thread. The more I study Rambus, the more convinced I am that it is a big mistake. I've given some of the reasons in
other posts and I don't have time to repeat them here but let me only add that the problems I have with Rambus are not due the the current problems but due to Rambus having long term problems with:
1. Cost
2. Performance
3. Reliability
Also, please realize that I am not saying that it is a bad design or the Rambus engineers are idiots. They are good engineers and I have several friends there and one of them used to work for me. I feel sorry for the Rambus design engineers. They have been asked to do one design that covers the entire memory market all the way from
a tiny system with one memory chip to a huge server with thousands of chips. It was a huge mistake to attempt this. The market is large enough to support several memory architectures each optimized to a segment and this is one reason the dram makers will outflank the one-size-fits-all Rambus.
Sorry that I'm rambling. Gotta go now.
|
