With the introduction of its 845D double-data-rate SDRAM chipset Monday, Intel Corp. closed the book on trying to dictate mainstream PC memory with Rambus.
Yes, Intel will still support Direct RDRAM for its high-end workstations and desktops, but this is a niche market. From now on, not only has Intel jumped on the DDR bandwagon full force, but the new convert is now trying to drive the DDR parade.
Intel also selected Direct Rambus as the memory to support its next generation network processor, to be introduced shortly.
After spurning DDR for years, Intel is now pushing the industry to ramp up DDR frequencies to 333MHz and 400MHz as rapidly as possible. Intel, with a few memory chip allies, also is hawking a new DDR design with on-chip impedance matching control for even faster speeds.
Quite a switch for the processor Goliath who only two years ago refused to put any DDR memory on its roadmap, claiming Rambus was the only next generation DRAM its processors would support.
Intel abandoned its last ditch Rambus effort -- a less expensive four-bank RDRAM design developed by Samsung Electronics Co. -- last fall by scuttling at the last minute its Tulloch chipset to support the new chip. The PC industry OEMs and suppliers correctly read that action as Intel leaving Rambus to swing in the wind.
(The four-bank RDRAM chip may still have a life as Silicon Integrated Systems (SiS) of Taiwan has taken out a Rambus license to build the chipset after Intel's rejection.)
In typical fashion, Intel simply ignores the several years memory turmoil it stirred in the PC market, as if it never happened. But OEMs and chip vendors who lived through the hectic times won't so easily forget.
Now that Intel is firmly on board the DDR train, almost everyone just wants to wash their hands of the "lost years" and get on with the business of regenerating the PC market with the memory that everyone now supports.
But Intel, which casts itself as an industry legend, should take some hits for writing probably one of the worst contracts in its history. Public documents at the Securities and Exchange Commission show that Intel agreed to not even discuss the possibility of coming out with a DDR chipset until 2003, in taking out the Rambus license.
Of course, that bar was obviously dropped in a revised licensing pact with Rambus, whose terms this time remained under lock and key. As the price of getting out of its original restrictive deal, Intel agreed to pay Rambus a one-term quarterly royalty fee, estimated by sources close to the deal at between $8 million-to-$10 million a quarter. Pocket change for Intel, and a small price to pay to get out of a very bad contract.
Intel also suffered a series of major technical goofs in its long series of efforts to promote Direct RDRAM as mainstream memory.
Remember the original 820 chipset fiasco, which left Dell Computer Corp. and other Intel OEM supporters with a potload of useless boards that had to be scrapped? That was followed by the red-faced Intel recall of faulty 820 and 840 chipsets and motherboards with a defective Memory Translator Hub that was supposed to allow the transition from SDRAM to RDRAM memory. That also blew up the projected Timna bargain-basement processor.
But now Intel appears to have its act together with a predictable roadmap linked firmly with DDR. There may be a lot less volatile upheavals to write about, but that is good news for the industry.