Click the download link above for a better look at ATI's Ruby demo.

Both new architectures from ATI and Nvidia offer incredible fill rate, vertex processing, and memory bandwidth gains, but ATI has chosen to target the Radeon X800 at the here and now by polishing its current design rather than by taking the risk of adopting new Shader Model 3.0 features. Nvidia has chosen to target the GeForce 6800 at the here, now, and more-distant future by allowing consumers to sacrifice convenience, a double-height card, and large power requirements for a more-robust feature set, including Shader Model 3.0. The new Radeon X800 does offer 2.x shader support, but it lacks the key SM 3.0 features of dynamic flow control, vertex texturing, and vertex streams.

Shader Model 3.0 is a collection of Microsoft DirectX 9.0c Pixel and Vertex Shader 3.0 features deemed necessary for next-generation 3D graphics. SM 3.0 hardware allows game developers to write longer shader programs with dynamic flow control, making the programmers' lives much easier and enabling more-efficient development time. Currently, developers are limited to short shader programs with limited branching capabilities. In SM 3.0, vertex shaders can access textures, allowing for true displacement mapping, which makes it possible for the graphics card to deform geometry in real time. Another SM 3.0 feature--geometry instancing--enables "batch" uploads of similar objects. Instead of spending valuable resources re-creating each object, geometry instancing allows the programmer to use a single base model and a large number of vertex streams that modify the model's appearance in each case.

The bad news is that the Radeon X800 isn't a Shader Model 3.0 chip. The good news is that we probably won't see a lot of SM 3.0-heavy games on the market until ATI finally does release an SM 3.0 part. Programmers can't risk spending a lot of development time on features that will only be enjoyed by a small percentage of the retail target market. Remember, we had to wait about a year and a half after the first DirectX 9.0 hardware hit the market before we saw high-profile 9.0 games, such as Far Cry.

The Radeon X800's Smartshader HD Pixel Shader 2.x engine offers more than enough graphical support for such high-profile game titles as Half-Life 2 and Doom 3. Actually, the previous-generation Radeon 9800 XT would be fine for both Valve and id's upcoming blockbusters. However, the new Radeon X800 doubles the pixel pipeline of the 9800 XT, adds 50 percent more triangle-processing power, and offers a few new features, including 3Dc compression, improved memory management, and a new temporal antialiasing technique.

Temporal Antialiasing.

Temporal antialising is a new antialiasing technique that alternates between two different sample patterns for each frame and allows the screen display's natural persistence rate to blend the images together. Temporal antialiasing may only sample at 2X, 4X, or 6X, but the effective antialiasing quality level will appear to be 4X, 8X, or 12X respectively.

Here is ATI's Optico villain with and without high-quality normal maps.

3Dc is ATI's new compression technology that's been designed specifically for normal maps. Games use normal maps to provide the illusion of 3D detail on a surface. The normal map adds information about the direction each pixel is facing to aid in creating realistic lighting effects, thus enabling the game to create even better 3D illusions on a flat surface. Unfortunately, normal maps also add pressure to the already taxed graphics memory subsystem. ATI's new 3Dc technology achieves up to 4-to-1 compression ratios, allowing developers to use more high-quality normal maps for better-looking 3D details.

The Radeon X800's increased pixel and vertex processing power will go a long way, since we're only just seeing Pixel Shader 2.0 games right now. ATI worked with design studio RhinoFX to create its new "Ruby" technology demo, which demonstrates the high level of 3D gaming that's enabled by the Radeon X800. The Ruby demo uses shaders extensively to create depth of field, hair rendering, soft shadows, and simulated subsurface-scattering effects.

The ninja is in focus, and the background is out of focus, which results in a depth-of-field effect.

Ruby is presented with hard shadows on the left and soft shadows on the right.

We tested our reference ATI X800 XT and Pro cards on a P4 Intel 3.4GHz Northwood processor with an ASUS P4C800-E Deluxe motherboard, 1GB (512MB x 2) of Samsung CL3 memory in a dual-channel configuration, and an Antec TruePower 480W power supply.

Nvidia managed to get us two new, previously unannounced GeForce 6800 boards--the GT and an unnamed "Extreme" edition--in time for inclusion in our testing. The 256MB GeForce 6800 GT will be a 16-pixel pipeline solution with a core clock of 350MHz and a 500MHz (1GHz effective) memory clock. The GeForce 6800 "Extreme" will have 16-pipes, a 450MHz core clock, and 256MB of 550MHz (1.1GHz effective) memory. Since this "Extreme" edition will be in limited release--based on however many "golden samples" (chips that are able to hit higher-than-average clock speeds) Nvidia is able to muster--we've tagged the 6800 "Extreme" with a "GS" moniker in the charts. We used a new 61.11 beta driver for use with the 6800 series cards, and we have disabled trilinear optimizations for all tests.

Our test resolutions are 1024x768 and 1600x1200, and we enable 4X antialiasing and 8X anisotropic filtering in our quality tests. When viewing the benchmark results, we suggest comparing cards in the same price range. The $499 Radeon X800 XT Platinum Edition should be compared to the GeForce 6800 Ultra, while the $399 Radeon X800 Pro should be compared to the GeForce 6800 GT.