RSS Feed

HCW Tech Blog

For the latest info on computer hardware, tech, news, video games, software tips, and Linux, check out our new improved front page: HCW Tech Blog

Reviewed by: Bryan Pizzuti [03.09.03]
Edited by: Carl Nelson

Card manufacturer: Powercolor
GPU Manufacturer: ATI

MSRP: $370
Street: $240-290

Discuss this article in the forum!


Card specifications:

GPU Frequency: 325 MHz
Memory: 128 MB 256 bit
Memory Speed: 310 MHz DDR (620 MHz effective)
Multi-monitor support: Yes
AGP 8x max
Vertex Shader 2.0 w/T&L program
Pixel shader 2.0
Fully DirectX9 compliant
SMOOTHVISION 2.0 AA/Ansio unit
TV-Out: 800x600 max
8 pixel pipelines
1 texture unit per pipe

DirectX9 is here

The primary focus (other than performance, of course) is the RADEON 9700 Pro's support for DirectX9 functions.  These include version 2.0 of the pixel and vertex shader standards, as well as floating point pixel pipelines.

So what ARE floating point pixel pipelines?  These are greater bandwidth, greater decimal accuracy pipelines, allowing for MUCH greater color accuracy and reproduction.  Cards before this one have always used 64-bit wide integer-only pipelines, but floating point allows for a greater degree of accuracy, and the floating point pipe itself is at least 96 bits. And yes, that's a big deal.  Now we can go beyond millions of colors, into millions of SHADES of millions of colors.  This will provide smoother transitions between colors in scenes, rather than having to use dithering and sharp gradients, which look much less realistic.  While the number of color and shade combinations isn't infinite, it's much too high for me to assign a number to, much less count. I'd show an example, but it's simply not possible using the 32 bit color settings you're probably using on your desktop. :) 

So what else is under the hood? Well, there are the vertex shader units.  That's right, PLURAL; not one unit, but 4 discrete, individual units, allowing more shader processes to run in parallel.  There are a total of 8 pixel pipelines, all floating point, with one texture unit each.  That one texture unit can apply 2 textures per pass, since the pipeline can loop back and put the pixel through the unit twice in a single pass.  This incurs a slight speed penalty when multitexturing, compared to having 2 units per pipeline.  However, GPU die space is at a premium with the wide pixel pipelines and 4 shader units; adding another 8 texture units would have been very difficult, and costly.  Tests will show if this method works or not, but the 9700 Pro should have the horsepower to do well either way.

ATI's own features

HyperZ III is an incremental upgrade to ATI's HyperZ technology, which provides z-culling, as well as lossless compression to increase memory bandwidth.  It now includes a memory bus similar to NVIDIA's "Crossbar" controller, providing multiple individual, narrower busses for more efficiency.  These 4 separate busses are composed of 64 bits apiece and add up to a total bus width of 256 bits (twice that of NVIDIA's memory busses, even on the upcoming GeForceFX 5800).  That memory runs at 620 MHZ (310 DDR) for a mind-blowing total of 19.2 GB/s bandwidth to feed the 325 MHz R300 GPU.  Another introduction to the HyperZ system is EarlyZ, which subdivides the Z-buffer down all the way to pixel level, thereby increasing accuracy when invoking z-culling routines.

Smoothvision has been updated to version 2.0, and now contains more efficient "jittered grid" multisampling methods of anti-aliasing through 6X.  Anisotropic filtering, for those familiar with early RADEON cards, has finally been fixed as well, and supports up to 16X sampling in both Quality and Performance modes.

TRUFORM is an exciting-sounding technology that is still awaiting wider implementation in games.  It provides a way of implementing a combination of polygon and curved-surface technology for increased realism (curved-surface models were used in Pixar's Toy Story movies, as an example).  Since current computer 3D technology is limited to polygons, what TRUFORM does is take the polygon data from the game, put it into the engine and convert it into a curved surface, at which time, transform and lighting operations are performed on it.  This actual curved surface is then converted BACK to polygons for display purposes, but too many more polygons that were originally defined.  In some ways, this is similar to the description of an antialiasing engine, but for polygons instead of pixels. The idea is to "smooth out the jaggies;" in this case caused by flat-surface polygons attempting to create a curved shape. TRUFORM is also intended to increase lighting accuracy for these surfaces. We'll reserve judgment until we can enable the engine in our real-world tests later and see if it works, how it looks, and how much performance drop there is.  Supposedly, all of this happens in the T&L/Shader engine to produce as little performance loss as possible, but there MUST be SOME performance loss, due to the polygons that are added by the process. Although TRUFORM was introduced with ATI's original asskicker, the Radeon 8500, it has only recently been implemented in game engines.

Here are some screenshots taken from a TRUFORM demo located on ATI's site, just as an example of what it's supposed to be able to do.  This first one is in normal viewing mode, with max "tessellation," or "truforming" enabled.  It's right below the non-enhanced dolphin, so you can see the difference.

Here is a screenshot showing wireframes.  This is in 16X mode.

Now, this is an ATI demo, rather than a game implementation.  This means that this model is fully optimized and designed for TRUFORM operations, whereas many games' models are not.  But it's an example of what TRUFORM is INTENDED to do, if it's implemented properly. 

Next Page: (3)