Hi jim kelley; Re the future of DDR as main memory. You've made an incredible number of unsubstantiated untrue comments about DDR and RDRAM. I'll go through them one by one and refute them with links...
Re: "It is too early to tell what the future of DDR as main memory is since there are no products yet available." This is true only in the sense that most people are not technically acquainted with the details to be absolutely totally certain what will happen, but here are some things to bear in mind:
DDR designs completed or in progress at ALL major suppliers of motherboards (page 3)
Closeups of the DDR section of various five different DDR motherboards: Pages 16,17,29,30,36
micronsemi.com
Re: "DDR does not scale well since it is essentially a clocking scheme and does not include as part of its specification control, address and data termination." Not true:
ami2.com
Re: "As the clock rate increases, DDR-II falls prey to the same issues of bus terminations already solved by RDRAM. Moreover, the bus has 4 times as many signal traces which compounds the termination problem. RDRAM is the correct approach to scaling in clock frequency." Not true. In fact, DDR is increasing in frequency at a rate much faster than RDRAM. Early DDR ran at 200/266 MHz, (data rate per pin). It's already shipping at 333, and sampling at 400. If RDRAM is the correct approach, how come it's only scaling from 800MHz to 1066MHz over its first two years of production? Looks like RDRAM has run up against a physical limit, while DDR has plenty of room to improve left in it.
Re: "Servers can afford the cost of a few extra layers of motherboard but desktop systems can not. The 820 is a 4 layer board." The implication is that DDR will require more layers than 4. Not true, as is well known through the industry. Take a look at these:
From VIA: DDR SDRAM is evolutionary
* One controller can support both SDR and DDR
* Same PCB (both module and MB) technology as SDR
micronsemi.com
For actual PCB drawings showing the 4-layers required, go to page 37 & 38:
micronsemi.com
For mom and pop who don't know their butt from an excellon drill tape:
PC2100 DDR can be implemented in a low-cost PCB (Standard PC100 motherboard technology (page 28)
DDR channel can be implemnted in a 2S 2P (four-layer) board (page 29)
Note: While the actual Micron Reference Motherboard design is six layers, you can clearly see that the memory interface was routed in just two signal layers. Also the Intel i820 Reference Motherboard was 6 layers, cost reduction comes later. Early DDR motherboards will (I predict) be 6 layer, but they will then migrate to 4 layer, just like any other engineering cost improvement. Until then, the extra cost of the two layers on a MB size PCB in large quantity will be well under $10 each, a tiny fraction of the difference between the price of DDR DIMMs and RDRAM RIMMs.
Re: "DDR offers a parallel raw bus that does not scale well as the clock rate increases." This is not true, as noted above. DDR has been increasing its clock rate at a far, far, far higher rate than RDRAM.
Re: "DDR SDRAM has been around since 1996 and has not found application in main memory despite many attempts over the years." This is not true. DDR samples have been around for years, but DDR mass production started in late '99. RDRAM mass production started about 6 months earlier. To help remind you of the facts, here are some links showing DRDRAM devices being sent to Intel at the same time that DDR was being sampled:
Samsung Ships Rambus Modules - DRDRAM devices to be tested by Intel, Compaq, and Dell August 17, 1998
usa.samsungsemi.com
Samsung Samples First JEDEC-Compliant DDR SDRAM August 17, 1998
usa.samsungsemi.com
Re: "Most the criticism of RAMBUS has centered on cost and realized performance with P3 systems. Both these issues will dissolve with the launch of the P4." Given that this statement is about an event which is firmly in the future, there is no way I can refute it directly. But people on this thread have suggested many, many, many times that Intel is a very savvy technical company, and that is why they made the brilliant decision of choosing Rambus for their memory interfaces. But Intel just announced that they are going to support PC133 for mainstream use of the P4, and rumors are flying that they will support DDR in 2H01. This move by Intel clearly implies that they do not think that RDRAM will be available in sufficient quantity (and therefore price) to support their P4 in 2H01. Since that is something like 9 months after the launch of the P4, it is therefore clear that they are afraid that the RDRAM issues will not dissolve before then.
Re: "The die area of the second generation of RDRAM devices will be within 5% of Sdram (as per Samsung). The rest of the cost equation arises out of packaging and economies of scale. Thus it is not really possible to assert that RDRAM will be more expensive than ordinary Sdram in 2001 and 2002." There are a number of severe inaccuracies in this statement. First, Samsung doesn't predict a reduction in the 5% extra cost for RDRAM until 2002, not 2001. Second, very few people are comparing RDRAM to SDRAM, (which it absolutely no hope at all of equaling the price of), but instead, they are more liberally comparing RDRAM to DDR. But DDR is already cheaper than RDRAM, and doesn't have to wait until 2002 for a cost reduction. Who knows how cheap DDR will be by then.
Re: "The main reason Intel announced an Sdram version of Willamette for the second half of 01 is to cut off AMD and VIA as well as the other chipset companies. As you recall VIA has been thumbing its nose and saying that it will build a DDR chipset for the P4 without a license. AMD has been touting DDR and Intel is intent on denying them any distinction in the market place." This is an example of magical thinking. It's kind of cute, actually. I'm going to post it around SI and see what other people think about it.
Re: "Finally, it is becoming increasingly clear with each new license agreement that RAMBUS will be collecting royalties from VIA..." While the thought is distasteful, this has nothing to do with the future of DDR, and I will refrain from commenting other than to note that this remains to be seen.
Re: "People are comparing DDR-II to RDRAM instead of comparing DDR-II to QRSL RDRAM. The comparisons have been picked to meet a political agenda. Still there are no DDR systems much less DDR-II systems on the market. Sure there are graphics cards that use DDR SGRAM but so what." The basic fact is that DDR-II is scaling to increased clock frequencies much faster than RDRAM is. As for QRSL, it is receiving very little support in the industry, as compared to DDR-II, which has the support of a lot more than just one memory company.
-- Carl |
