To: Bilow who wrote (42340 ) 5/16/2000 1:16:00 PM From: Bilow Read Replies (1) | Respond to of 93625
Re die areas. More sham engineering from the secret Rambus hype site, Dramreview... The article is more reasonable than a good portion of the people posting on this thread. For instance, they agree that DDR will only require 6 layer boards, while posters here have misquoted Intel to the tune of 12 layers:Both RDRAM and DDR memory systems will require 6 layer boards to route signals and fit the modules into the smallest possible form factor. dramreview.com They go on to suggest that the tie breaking advantage to RDRAM in cost is due to the cost of the North Bridge controller chip. They site an AMD study which gives the pad limited areas for the controllers as 107mm2 for RDRAM vs 133mm2 for DDR:dramreview.com Of course, being a pump and dump site, they don't give a context for the above information. But lets take a close look at these figures... First of all, if a chip area has to increase due to pad limiting, this means that the chip will end up with "white space", regions on the chip with no circuitry. There is much less worry about defects in the white space, so there is no reduction in yields in terms of percentage good die. Instead, there is only a reduction in die per wafer, and it is roughly proportional to the die area. In this case, the DDR area is 15% larger, so the chips will cost 15% more. But how much do North Bridge chips cost? The chip is typically the largest on a motherboard (without CPU), and since the motherboard costs around $125 or so, you can guess that the cost of goods sold on the main chip has to be around $30. The alleged DDR adder for the North Bridge is then about $4.50. Of course the RDRAM controller carries with it a royalty cost, as do all the memory chips, and given that the memory chips are each about 100mm2, and that there are 8-12 of them, the royalty costs of RDRAM about match the alleged die controller cost of DDR. But RDRAM also means bigger memory die area, and since there are a lot more memory chips than there are controllers, the memory area dominates the calculation. Even including a 5% increase in memory die size, with 10 memory chips, the die penalty is 50mm2, about three times the 16mm2 that they are going on about. Just more sham engineering from "r"dramreview, a site that fakes an unbiased attitude. And where are those GeForce2 products in their technology list, anyway? I predict that these guys are going to drop that list sometime late this summer, when it becomes clear that DDR is here to stay. The primary reason is that the format they chose, to list the users of chipsets, is one that will be highly disadvantageous to RDRAM in the near future. Rambus' design wins are concentrated in a small number of companies that are more vertically oriented, while DDR designs go into multiple products but with smaller total unit volumes (so far). As an example, look at their small system totals. DDR already outnumbers RDRAM by 6 to one, and they haven't even included the GeForce2 cards now widely available. -- Carl