Dueling graphics juggernauts trade announcement, improvement jabs
By Brian Dipert -- EDN, July 8, 2004
With Nvidia's latest graphics flagship now disrobed and in feisty fighting shape, you had to know that ATI Technologies' counterpunch wouldn't be too long in arriving (see “Graphics advancements span PCs to cell phones”). ATI's mainstream X800 Pro and high-end X800 XT Platinum Edition devices might strike you, at first glance, as nearly identical to their Nvidia competitors, the GeForce 6800 and 6800 Ultra.
Both vendors' mainstream parts have six vertex-shader units and 12 pixel pipelines in four-pipeline-per-group granularity with two pixel-shader units per pixel pipeline. Both high-end parts bump the pixel pipeline count to 16 and interface to a 550-MHz GDDR-3 (graphics-DDR, third-generation) SDRAM over a 256-bit memory bus with 1.1-Gbps per-pin peak bandwidth driven by a four-way crossbar memory controller.
ATI and Nvidia claim that their devices hardware-accelerate MPEG and Windows Media Video decoding along with MPEG encoding, although ATI accomplishes the tasks in multifunction shaders, whereas Nvidia employs a dedicated video processor. Peer closer, though, and some significant differences emerge, differences that, at least in the near term, will likely play in ATI's favor.
ATI has consciously chosen to not yet support DirectX 9.0c's Shader Model 3 instructions and their 32-bit precision, instead continuing to rely on the claimed “more-than-good-enough-for-now” 24-bit precision that the company pioneered in the previous generation Radeon 9700 family and Microsoft supported in DirectX 9.0b and Shader Model 2. This decision has enabled ATI to construct its X800 devices from “only” 160 million transistors versus 222 million, or 39% more, for Nvidia. Couple that fact with the ATI devices' low-K, 0.13-micron fabrication process, and the result is that the X800 XT Platinum Edition runs at a 520-MHz core frequency—30% faster than the Nvidia Ge-Force 6800 Ultra—and burns only an estimated 65W of power when running Futuremark's (www.futuremark.com) 3DMark03 benchmark suite—60% of the 6800 Ultra's estimated power draw.
Citing the high cost of 4 Mbit×32-bit memories required to construct 128-Mbyte frame buffers with a 256-bit aggregate bus interface, ATI plans to offer no X800 board variants that employ them. The X800 Pro, running at a 475-MHz core frequency and with 256 Mbytes of 450-MHz DDR SDRAM, costs $399, and the company is now shipping it. The premium X800 XT Platinum Edition costs $499 and should have hit store shelves on May 21.
At the 11th hour, Nvidia pulled two additional GeForce 6800 SKUs (stock-keeping units) from its hat in a competitive response. The $399 6800 GT runs at 350-MHz core and 500-MHz memory speeds, with a 256-Mbyte frame buffer. (Recall that the $299 conventional 6800 board offers only a 128-Mbyte frame buffer and reportedly runs at 325-MHz core and 350-MHz memory speeds.) Nvidia also announced the 6800 Ultra Extreme, with unspecified higher-than-6800 Ultra pricing and clock rates.
Nvidia's first wave of GeForce 6xxx desktop graphics introductions is now out of the way, and the company has turned its attention to preparing the industry for upcoming mobile variants of the architecture. At the same time, Nvidia is responding to those of you who want to upgrade your notebook PCs' graphics subsystems but without locking yourself into the proprietary-graphics upgrade modules from companies such as Alienware (www.alienware.com) and Dell (www.dell.com). Nvidia's proposed MXM (Mobile PCI Express Module) standard, which it hopes that other graphics vendors will also embrace, comes in three module sizes commensurate with graphic-chip dimensions and frame-buffer densities. It builds on the proven, cost-reduced SODIMM socket, bumping up the connector's contact count to 230 pins and beefing up its sturdiness and current-carrying capability to reflect the larger, heavier, and hungrier graphics modules (Picture).
What does Nvidia plan to do with all those contacts? Among other things, they’ll provide sufficient power supply flexibility to feed numerous graphics- and memory-chip subsystems with varying voltage and tolerance needs. Also, MXM handles a mind-boggling array of simultaneous displays: dual-link, 24-bit LVDS, VGA, S-Video, composite video, component video, and two independent DVI ports.
ATI Technologies, 1-905-882-2600, www.ati.com.
Nvidia, 1-408-486-2000, www.nvidia.com.
