Rival graphics memory specs spread confusion

Rival graphics memory specs spread confusion
BEIJING — The acrimony of memory industry politicking was on display at the JEDEX Beijing Conference Thursday (Oct. 24) as dissatisfaction brewed over the characterization and control of emerging graphics DRAM standards.

On the sidelines of the JEDEX conference here, memory industry insiders complained about growing pressure from graphics chip suppliers that want to tailor upcoming DRAM to meet the needs of their high-end, low-volume graphics chips. In some cases, sources said, the top two graphics chip suppliers have tried to get manufacturers to include proprietary features in their memories that would preclude competitors from using the chips, further splintering an increasingly divided memory market.

The debate stems from graphics chip maker ATI Technologies Inc.'s plans announced earlier this month regarding GDDR-3, an alternative to DDR-2 memory. ATI said it had hammered out a new graphics DRAM spec with archrival Nvidia Corp., and said memory manufacturers had agreed to implement it.

In short, the GDDR-3 spec defines key changes in the memory's clocking mechanism and I/O scheme so that it is better-suited for point-to-point links to a graphic controller; this approach differs from DRAMs designed to send and receive data over a bus. The first GDDR-3 devices should arrive next May or June, with speeds starting at 500 MHz and throughput of 1Gbits/second per pin. Later versions will scale to 1.4 to 1.6 Gbits/s.

Problem is, what ATI calls GDDR-3 is quite similar to what Nvidia, Samsung and others are calling GDDR-2, samples of which were shown at the Beijing conference. "What it really comes down to is that what they are calling graphics DDR-2, the 1-Gbit/second part, was a spec put together by Nvidia," said Kevin Ryan, a fellow at Micron Technology Inc. "The GDDR-3 spec was worked on with ATI. So really you have an Nvidia spec out there that goes up to 1 Gbit/s and an ATI spec out there that goes up to 1.4 Gbits/s. So ATI felt that because it went 40 percent faster, it needed to be called 3 and not 2."

Marketing gimmick?

Others are outright critical of ATI for overselling the device by naming it GDDR-3, which they say blurs the distinction between what is essentially a DDR-2 part that's been optimized for graphics and the still-undefined DDR-3 standard. "It's a marketing gimmick just to confuse people," said Jon Kang, senior vice president at Samsung Electronics Co. Ltd., a top supplier of graphics DRAM that also worked on the spec. "It's also the son trying to tell the father what to do," he added, implying that ATI is trying to play catch-up with Nvidia's graphics spec, which is expected to be implemented in parts that will hit the market first.

Another Samsung official said that further modification to the chip interface will scale the company's current GDDR-2 device up to 1.4 or 1.5G bits/s.

If the companies carry through with their plans, observers said there will be two new types of graphics memories on the market next year — GDDR-3, which will be used by ATI, and another probably called GDDR-2, used by Nvidia. "They are fundamentally the same thing," said Desi Rhoden, chairman of Jedec, which oversees the development of chip standards. "I'm sure the next company that comes up with a proposal will call it GDDR-4, or 5 or 6."

Jedec will likely complete its specification for graphics memory based on DDR-2 in December, and may issue another name for it at that time, Rhoden said.