So far, the companies are hedging their bets and say they won't rule out the adoption of both memory types. However, some smaller graphics companies that lack the resources of their larger competitors may be forced to pick a winner early on.

Nvidia, based in Santa Clara, Calif., is the first vendor anywhere to use GDDR-II in production volumes by designing the memory into its high-end GeForce FX graphics card. GDDR-II is the nomenclature assigned to the graphics-specific variant of next-generation double-data-rate SDRAM (DDR-II), and is so named because it has been optimized for graphics applications through the inclusion of point-to-point signaling.

Samsung Electronics Co. Ltd. so far is the sole source for the 500-MHz, 1-Gbit/s data rate GDDR-II memory. Infineon Technologies AG and Elpida Memory Inc. are expected to sample GDDR-II in the first quarter of 2003, while Hynix Semiconductor Inc. is slated to introduce a similar chip in April of next year.

Micron Technology Inc. is virtually alone among top-tier DRAM makers in opting not to manufacture the GDDR-II chips. Instead, the company will begin marketing GDDR-III samples in the second quarter of next year, according to Terry Lee, executive director of advanced and strategic marketing.

"The GDDR-II market will be very short-lived," Lee said. "The transition will be faster if GDDR-III chipsets can be designed to be backwardly compatible with GDDR-II."

GDDR-III is also in the product roadmaps of Elpida, Hynix, and Infineon, with initial production expected in the second half of 2003. Taiwan's Nanya Technology Corp. said it has no plans to make either graphics memory.

A race developing

Jon Peddie, of Jon Peddie Research, Tiburon, Calif., foresees a race developing between the two DRAM versions to determine which will be the first to achieve cost-effective yields and attractive pricing.

"It's definitely going to be a fragmented memory market for graphics next year," Peddie said.

Matthew Godfrey, a DRAM analyst at Semico Research Corp., Phoenix, Arizona, said the mixed market will complicate the supply picture for manufacturers of graphics processors that rely on advanced memories.

"And confusion in the market equals customer rejection," said Jim Sogas, vice president of marketing at Elpida's U.S. subsidiary in Santa Clara.

Neil Trevett, senior vice president of market development for graphics hardware designer 3DLabs, Milpitas, Calif., agreed. GDDR-III is the first choice for 3DLabs' next graphics processor coming in 2003, but the company has a fallback position that includes the utilization of GDDR-II.

"There's a potential schism in the graphics memory market," Trevett said. "We would much prefer for the memory guys to be more unified."

Though Nvidia and ATI each has decided which DRAM horse it will ride, executives at both companies said customer demand ultimately will determine which memory type receives their full support.

Nvidia, for instance, is using GDDR-II for its GeForce FX processor. "We'll be ready to ship a product with GDDR-III when it's ready. It all depends whether GDDR-III turns out to be a faster memory than GDDR-II," said Bryn Young, director of memory sales and operations. "Otherwise, it's not that interesting."

Trident Microsystems Inc., Mountain View, Calif., also is going with GDDR-II, but engineers are evaluating GDDR-III should the graphics vendor decide to pursue the option, said Le Nguyen, vice president of marketing. "It depends on which memory is faster and has the higher bandwidth," he said. "There is no such thing as having enough bandwidth in graphics."

Even ATI, Thornhill, Ontario, which spearheaded an industry effort to design GDDR-III, has demonstrated that it is able to use GDDR-II to support its Radeon 9700 and Radeon 9500 processors, said Joe Macri, director of engineering. "GDDR-II is a transitional memory until GDDR-III comes to market," he said.

Further confusion

Creating further confusion is the fact that GDDR-II is itself fragmented into three versions, each of which uses different register sets. Elpida's Sogas said there has been such a rush to market by DRAM manufacturers that different GDDR-II designs were finalized before a single JEDEC standard was drafted.

"We have a little bit of a mess on our hands," acknowledged ATI's Macri, who is also chairman of the DRAM memory committee of the JEDEC Solid State Technology Association, Arlington, Va. "But JEDEC is working to draft a single GDDR-II standard with a common specification on registers. We expect to complete the standard by September 2003."

In the meantime, Nvidia and ATI both claim that their memory controllers can handle all three GDDR-II register sets, allowing them to use multiple suppliers.

As the effort to rationalize the standard continues, however, the market will exert its own force. Samsung, for example, will have produced a huge quantity of its GDDR-II chips by September, which will render its design a de facto standard, according to Mueez Deen, director of marketing for DRAM graphics memory at Samsung Semiconductor Inc., San Jose.

Deen added that the fast pace of the graphics processor market compresses memory products into a six- to 12-month lifespan. In fact, Deen said Samsung already is developing an entirely new advanced graphics DRAM design to succeed both GDDR-II and GDDR-III, with first samples slated for 2003.

Sogas said he expects GDDR-II and GDDR-III eventually to find complementary roles.

"GDDR-II will likely be less costly, but GDDR-III will likely be higher density," he said.

Following behind is DDR-I, which vendors are still driving to higher clock speeds. Hynix last week introduced a 500-MHz, 1-Gbit/s graphics DDR-I that the company claimed is the fastest in the industry. That is actually a higher clock frequency than some versions of GDDR-II and equals the clock rate of the GDDR-II used in the Nvidia GeForce FX card.