Why Intel and Rambus are so close

Analysis Many people have wondered why it has taken Intel just so long to say it will support PC-133, even though its acceptance by the chip giant is hedged about with many a caveat. Interest in the relationship between Intel and Rambus has been fuelled not only by the small matter of endorsing PC-133, but also by the fact that most of its big hardware customers, including Dell, HP and IBM, have now said that they will use DDR memory in future servers. Rambus, too, does not seem to play much of a part in the future of Intel's flagship processor, the 64-bit Merced, which the company demoed at its Intel Developer Forum last week. And that interest is once more compounded by the news that Intel engineers have been forced to abandon implementing Rambus in the Carmel quad chipset, and have instead turned to Reliance for a solution. Last week, Intel wheeled in seven memory manufacturers to the press at IDF and announced the formation of a Rambus implementors forum, to further encourage the adoption of the technology. It's no secret that the large memory manufacturers have never had much of a soft spot for Rambus technology. To manufacture product, they had to pay for a licence from Rambus, and, thereafter, pay the company a royalty fee. The memory manufacturers already felt their margins were squeezed enough, and resented being "bumped" by Intel and Rambus into the technology. At the same time, they, and the PC vendors, were not persuaded that Rambus could perform to the promised level and felt, rightly or wrongly, that the cost premium of Direct Rambus in line modules (RIMMs), was too high. To hedge their bets, the memory companies were forced to support both Rambus and competing technologies such as PC-133 and DDR memory, all the while conscious of the fact that you don't play fast and loose with the Great Satan of Chips. It was UK executive Jamie Minotto, when he headed up Tandem 10 years ago, who said of Intel: "You don't tread on the tail of a tiger." It now emerges that the i820 Camino chipset, which uses Rambus memory, may now actually underperform compared to the BX chipset using SDRAM. (For some technical information on these questions, Tom's Hardware Page has analysed the chipsets. Go here and here for his conclusions.) Some answers to the question of just why Intel is so closely linked to Rambus are, in fact, to be found in public documents. In the SEC archives, there is a 10-K filing by Rambus which could give the clue to the relationship. In an AMENDMENT NO. 1 TO SEMICONDUCTOR TECHNOLOGY LICENSE AGREEMENT is the following agreement: "Intel will use its continuing best efforts in marketing, public relations, and engineering to make the Rambus-D DRAM the primary DRAM for PC main memory applications through December 31, 2002; and (b)Intel will communicate to the top (10) DRAM manufacturers, Intel's intention to support the Rambus-D Interface Technology in its integrated circuits for low end workstation, performance desktop, and basic PC platforms." The reason for Intel agreeing to these terms is likely to be the existence of patents which hold the two companies in a tight embrace. But there's more. In the same section is the following statement: "In January 1997 the Company granted a warrant to Intel Corporation for the purchase of 1,000,000 shares of Rambus common stock at an exercise price of $10.00 per share. The warrant will become exercisable only upon the achievement of certain milestones by Intel, which will result in a charge to the statement of operations at the time of achievement of the milestones based on the fair value of the warrant." That deal is now worth between $80 million and $100 million, a sizeable carrot which is worth a lot to Intel's bottom line. Intel is obviously deeply committed to its deal with Rambus. You only have to look at the notebook roadmap we published over the weekend to realise that. Here, it's worth looking at the PC700/PC800 note in the slide we published. PC700 is a new Rambus phenomenon. PC800 is 400MHz Rambus while PC700 is the 355MHz frequency, which is a fall-back position. It is also difficult to see where applications and processors that take advantage of 1.6Gb/s bandwidth will come from, while the AGP 4x bandwidth will be mostly unused. It makes more sense here to place large graphics buffers on cards and not use AGP at all. Intel may well have found itself in a hole, and, what's more an expensive and deep hole over the whole, vexed subject. ® See also Intel abandons server Rambus efforts Seven Dramurai™ ride two memory horses at once Does BX chipset with SDRAM outperform i820 with Rambus? Intel move to PC-133 mere lip service Intel pushes Rambus hard