ANALYSIS: Rambus headed for workstation niche

ITworld.com –

Although its streamlined bit-moving scheme is of little interest to most computer buyers, Rambus is front and center when next-generation computing architectures are under discussion. Until recently, it was thought that Rambus would be a big part of future computers that used Intel microprocessors. Word of trouble with a low-end Intel chip known as Timna and a series of chip-related license deals have put Rambus in the news.

Ultimately, various chipmakers and systems makers will decide what types of computer land on the desktops and in the computer rooms of IT shops everywhere. At the moment, it appears that Rambus, once seemingly destined for wide use, may be relegated to the high-end workstation market, leaving the low-end desktop and all-important server markets to alternative memory architectures.

As most buyers know, CPUs have vaulted ahead in speed, but the memory associated with those CPUs is often a performance bottleneck. Chip designers foresaw this several years ago, and began experimenting with memory bus innovations such as Rambus.

Future computers often will have "Intel inside," but whether they use Rambus is not yet clear. Some observers suggest that Rambus is a good fit where multimedia and graphics use is intense, but, unless streaming media quickly enters the mainstream, that support may not be needed right away.

In any case, Rambus is often found in the news these days, usually in a none-too- favorable light. Generally the company is discussed in terms of suits and countersuits -- or settlements -- with the big names in the memory chip business. It has been suggested that the big chip makers bridle as a result of dealings with Rambus, which is more a high-tech think tank than a traditional chip house. Rambus designs chips and establishes technology patents, but doesn't fabricate its own designs.

Reports in recent days have cast doubt upon Intel's endorsement of Rambus.

Intel's decision in the mid-1990s to use the Rambus DRAM (R-DRAM) architecture as the technology of choice for deploying future high-speed designs was a crowning achievement for Rambus. It highlighted what was until then a foreign method for moving data on and off fast processors. Rambus was expected to become a common site on the computer terrain based on its choice by the 300-pound gorilla of desktop computing.

But Intel and others have found Rambus technology difficult and expensive to implement. Anti-Rambus rumblings grew louder when Intel CEO Craig Barrett told the Financial Times of London that Intel's decision to go with Rambus memory was a mistake. Intel officials subsequently said Barrett was referring to Intel's Timna processor, a Rambus-compatible design aimed at low-end computer systems. Timna was cancelled before its launch due to problems with a translation hub that would have allowed the Timna processor to operate with less expensive synchronous DRAM (S- DRAM).

Intel is publicly committed to keep working with Rambus, albeit solely for high-end workstation designs.

An Intel spokesman said that Intel decided to drop support for the R-DRAM in its so- called "value line" of Timna products owing largely to the high costs of the technology, and will instead use alternative double data rate DRAM (DDR DRAM) technology.

Intel ran into problems with its Timna project, which had been aiming to create a low-cost PC by integrating most PC functionality onto a single chip, when the company tried to add support for S-DRAM. Memory makers and systems houses wanted to use S-DRAM as an alternative and Intel had created a chip to translate the signal from the chipset design to R-DRAM mode. But the translation was somewhat flawed, and since Intel realized that the market would be gone by the time it could correct the problem, the company decided to terminate Timna. Before the company could get a translation hub chip to work correctly, it was felt, the market opportunity would have come and gone.

There may be other issues associated with the use of Rambus technology as well. Kevin Krewell, an analyst with MicroDesign Resources, said, "Rambus does not scale as easily and you cannot add as many modules on a board. So from that point of view, all of the OEMs have told Intel that they need to focus on S-DRAM and follow on with the DDR S-DRAM."

Krewell added, "Rambus handles packetized traffic [more efficiently] and S- DRAM handles [greater amounts of] DRAM better. Rambus is good for graphics and media processing and S-DRAM is better for all-around general-purpose use."

Dean McCarron, an analyst with Mercury Research, does not agree that Rambus is necessarily better for graphics. He noted that Rambus has miniature caches that can deliver data more quickly, but the downside is that you may have to access data in big chunks in order to realize improved performance.

McCarron said the primary difference between DDR DRAM, S-DRAM, and R-DRAM lies in the electrical interface to the circuit. He notes that the manufacturers assert that DDR DRAM will be easier to build and has fewer royalty costs, but added, "We are still early in the market for DDR to determine if that is true."

Said Hemant Dhulla, Intel's marketing director for enterprise chipsets: "One of the reasons why we are utilizing S-DRAM in the mainstream desktop space is the lower cost of the platform. As one moves up from midrange workstations, we continue to believe R- DRAM and dual-memory channel [connecting] chipsets are the better solution. [Ed. note: DDR-DRAM is a synchronous format. It evolved from the earliest S-DRAM designs. The terms S-DRAM and DDR-DRAM are sometimes used interchangeably.]

R-DRAM memory incurs a premium; this is an issue in the server market, where using less expensive memory is more crucial than it is in high-end workstations. Dhulla admits that DDR-DRAM may be a better play here.

"In the server market, the typical server has a boatload of memory associated with it. DDR allows computer designers to build up the density of memory. Lower cost and higher performance can be achieved with interleaved memory subsystems designs," he said.

In the low-margin mainstream systems business, a royalty charge can upset vendors. In fact, royalty issues are among the most niggling for Rambus. For big semiconductor manufacturers, and for board or systems makers further down the food chain, the obligation to pay royalties on every Rambus-enabled computer shipped has been a stubborn requirement all along.[Rep.note: Rambus officials were unavailable for comment.]

The royalty issue does not go away, even if Rambus were to disappear, however. Rambus holds patents that relate to competitive next-generation architectures such as S- DRAM and DDR-DRAM, the latter highly touted by Intel competitor AMD. Both of these technology standards emerged in the wake of Rambus technology, forged by large semiconductor makers that had grown slow to innovate in their work on memory-exchange protocols. But Intel's Dhulla suggests that patent portfolios are part and parcel of the semiconductor world, and those relating to DDR-DRAM will not prove to be inhibitors.

Includes material from IDG News Service sources.

Insider: How the basic tech behind the Internet works
Join the discussion
Be the first to comment on this article. Our Commenting Policies