Jase,
I think the ultimate irony may be that Intel tried to get the RDRAM pump primed with the Pentium III, so it would be hitting its stride with the Pentium 4 (which would not use any other memory), but by doing so, essentially destroyed RDRAM because it developed such a bad reputation. The Pentium III always had SDRAM support, whether from the aging 440BX, the castrated 810 with onboard video, or the fairly decent 815 later on. So, it allowed a transition while the 820 chipset (and the much lower volume, but excellent 840) slowly gained more market share and RDRAM ramped up production. In the event, the 820 showed poor performance and the memory was much more expensive, leaving RDRAM with a terrible reputation which was undeserved. By the time the Pentium 4 came out the reputation of RDRAM was so tarnished even the superlative performance it offered on that platform did nothing to mitigate its demise. Also, because it failed to attract buyers on the Pentium III line, the price was still very high as the ramping Intel had anticipated never happened.
I wouldn't exactly say it was Intel copying AMD, so much as Intel having to follow the marketplace which hated RDRAM because of the issues with the P6 platform. Incidentally, the performance on the 840 was actually quite good, but that was a workstation class chipset and never sold well, and the memory was (and actually still is) very expensive. I still use these boards a lot, as they are very reliable and have a lot of useful features, and I can tell you that even today, 1 GB of RDRAM, even on eBay, costs around $200 used, or $250 new.
I don't know that AMD would want to use XDR in all their machines, and it may be that DDR2/3 will perform better in some applications while XDR might be better for others. You might see it just in a specialized area, maybe servers, maybe extreme game machines, maybe workstation class machines. I don't know, really, but AMD licensing this could indicate they have found one or more areas that this memory offers enough of an advantage to create a product based on it. It may indicate much less of course, too.
Sun's new processor could perhaps indicate a new trend where companies offer more specialized products for different markets. For the market it is in, nothing can touch that processor, but if you take it out of its niche, it performs very poorly compared to less specialized designs. So, I wouldn't be too surprised if other companies create platforms that are particularly well suited for a specific type of application or applications, and not release one product for them all. I think, in particular, Sun's introduction will necessitate a response from other companies since I doubt they want to cede that market to Sun, and they currently have nothing even close to compete with them.
We may see the 486 come back in some form in the future, with several of these per die. They would be simple and run applications with massive amounts of threads very well, be very cheap to make, and use very little power (remember the 486 didn't use decoders and only had a five stage pipeline). It would be really strange, but not nearly as unlikely as it was before the Niagara.
[Posted by: TA152H | Date: 01/04/06 04:33:25 PM]