Intel chipset volte face will hit Rambus hard
Amador hammer of the Dramurai
Analysis Amador City, close to the strangely named Dry Town in a wine-growing area of California, is the code name of choice for Intel's excursion into the wonderful world of SDRAM, PC-133 and double data rate (DDR) memory next year. Information we received yesterday from two separate semiconductor analysts in the US, suggest that the Amador chipset, which will use PC-133 memory, signifies a further shift by pragmatic Intel away from Rambus. Meanwhile, one PC manufacturer we have talked to this morning is saying that Intel has just confirmed to him it will launch i820 Rambus boards formally on the 15th of November. Some mobo manufacturers, at least, will ship four RIMM versions of the i820 He said: "Our mainboard vendor on this project is still saying they will ship a 2+2 solution. They are happy with their tests and I have to say we have seen no problems on this solution. From a technical point of view Rambus appears to offer nothing in the consumer market with benefits only in professional workstation or Server systems. So a four slot DIMM solution would be better. However, the Intel design is only two DIMMs" Yesterday it seemed as if Intel might slip until the end of November on its solution, but it is determined to make a splash at Comdex/Fall. That is not the same as shipping, of course. In the second half of next year, we understand, Intel will also promote an extension of this chipset which will support double data rate (DDR) memory. That is a volte face for Intel but most observers believe that the company has little choice. We reported in the summer that major PC manufacturers, including Compaq, HP and IBM, were less than pleased with the technical results they were getting from Rambus technology. At the Computex trade show in Taipei, Mosel-Vitelic was outspoken about the Rambus platform and pointed to specific technical problems with implementing the memory solution. We understand that many of the memory manufacturers feel the same way, but are timid about expressing their fears. One semiconductor analyst told us yesterday that to some extent the problem rested on their doorsteps. The manufacturers disliked having to pay Rambus royalties in the first place, as it cut into their already skeletonic margins. The analyst said that as memory is such a cutthroat, commodity based business, any fluctuations or uncertainty in these markets made their quest for profitability that much the harder. DDR, however, is not the perfect technology to propel servers into the next decade and lacks the scaleability, or the potential scaleability of the Rambus platform. That, the analyst said, made it imperative on the manufacturers to come up with a technological solution of their own, something they had signally failed to do. Intel has plunged a great deal of its loose dollars into the Rambus project. Not only does it have a share in the company itself, but it spent half a billion dollars encouraging Micron to put more efforts into the Rambus initiative. Our information is that Intel initially offered them a billion bucks, but Micron turned this amount down because it felt it would be too much in the chip giant's pocket. Like many of the other Dramurai, Micron has to hedge its bets and support alternatives, just in case they turn out to be the industry standard. Samsung, too, received a sizeable investment from Intel to promote the Rambus platform. The logic Samsung Semi executives exercised here was coloured by the up-and-coming merger of Hyundai and LG Semi, we understand. Samsung wanted to be seen as the Korean manufacturer of excellence. The industry sensed the first whiff of Rambus trouble as early as February this year, when Intel was forced to delay the introduction of its i820 Camino chipset because of unspecified problems, but probably Rambus related. We're still waiting to see that chipset. We also smelt a Rambus rat at Intel's last Developer Forum when it became apparent that the Merced-Itanic project would not use Rambus memory either. This is all very embarrassing for Intel but there's no doubt whatever that Rambus will continue to be profitable. It's not dependent on the x.86 platform alone and does have promise for the future. The whole episode should serve to remind us just how interlinked the whole industry is, and how dependent one technology is on another. ® Mosel-Vitelic announces PC 133 support Computex 99 coverage Why Intel and Rambus are so close Intel abandons server Rambus efforts
Sponsored: Are DLP and DTP still an issue?