![[H]ardOCP](/images/blank.gif){kind=small}
- Date:
- Thursday , July 18, 2002
- Author:
- Kyle Bennett
- Google +1
Radeon 9700 White Paper
A very well written white paper covering the major internal workings of the DX9 compliant ATi R300 Visual Processing Unit along with our thoughts.
Memory Interface:
The RADEON 9700 incorporates a new high-performance 256-bit DDR memory interface, capable of providing over 20 GB/sec of graphics memory bandwidth. It includes four independent 64-bit memory channels, each of which can be simultaneously writing data to memory, or reading data back into the graphics processor. Sophisticated sequencer logic ensures that all four channels are being utilized for maximum efficiency.
The RADEON 9700 memory interface is also designed to support next-generation DDR-II memory technology. Expected to become available in early 2003, it will enable a 50% increase in memory performance.
Does the graphic above look familiar? It should as it is nearly identical to what we saw NVIDIA bring to the market with their latest GPUs. Even though NVIDIA made it familiar, Silicon Graphics is responsible for pioneering similar technology. This is not to dog on ATI for lack of originality, but performance dictates what is the best way to go in a situation like this and certainly using four independent 64-bit channels is not a bad idea, no matter who did it first. Once we crossed the threshold into 32MB memory buffers it certainly behooves you to be able to use the entire memory buffer in a more efficient manner than possibly wasting the entire 128MB (or 256MB) for a small texture. Also, as we see memory buffers getting bigger and bigger, we can surely be certain that we will see memory controllers getting more and more complex. Hmm, does eight controllers being used on a 256MB or 512MB RAM buffer sound excessive? Then again, maybe the advent of affordable DDRII will keep that from becoming a reality so quickly.
Vertex Processing Engine:
A 3D scene is composed of interlocking groups of triangles that make up all visible surfaces. By performing mathematical operations on the vertices at the corners of each triangle, the vertex processing engine can place, orient, animate, color, and light every object and surface that needs to be drawn. The process is controlled by small programs called vertex shaders that are uploaded to the graphics chip and executed by the vertex processing engine.
The RADEON 9700 features the most powerful vertex processing engine ever designed, incorporating four parallel vertex shader pipelines. Coupled with a highly optimized triangle setup engine, the RADEON 9700 is the first chip capable of processing one vertex and one triangle per clock cycle. It is also the first chip to fully implement the 2.0 vertex shader specification introduced in DirectX 9.0.
The one triangle per clock is simply a breakthrough number unlike anything we have seen on the desktop before. In terms of raw numbers this gives the Radeon 9700 roughly 2.5 times the triangle drawing power of the current GF4 Ti4600. It will certainly be fun to watch the benchmark wars as people start overclocking the 9700. Keep in mind that this VPU is still built on a .15 micron architecture. A transition to .13 could certainly supply a monumental boost in performance. I am not sure if it is likely but we might see an “Ultra” down the line as .13 micron fabrication is hashed out by chip builders. Then again, that is total speculation on my part.
out
Each vertex shader pipeline in the RADEON 9700 is designed to handle vector and scalar operations simultaneously. Vector operations work on values composed of multiple components, such as 3D co-ordinates (x,y & z components) and color (red, green, and blue components). Scalar operations work on values with just a single component. Since vertex shaders typically include a mixture of vector and scalar operations, this optimization can improve processing speed by up to 100%.
While it is really hard to get a handle on how vertex shaders are working for us many times, you can think of them at the backbone of providing smooth animations with proper lighting first and foremost in your gaming experience. There are a couple of instances in which we can see the power of the quad-vertex shader engine, most notable in 3DMark 2001 and we will talk about that in our subjective article today.
The incredible performance of the RADEON 9700 vertex processing engine is required to enable the next generation of interactive 3D experiences with realistic, cinema-quality visuals.
The Vertex Processing Engine of the RADEON 9700 also includes support for TRUFORM 2.0, the latest incarnation of ATI’s innovative higher order surface technology. It is designed to smooth out the curved surfaces of 3D characters, objects, and terrain by increasing the polygon count through a process called “tessellation”. By taking advantage of the massive vertex processing power of the RADEON 9700, it delivers more natural-looking 3D scenes without requiring any changes to existing artwork. To learn more about TRUFORM technology, please visit www.ati.com to download the white paper or view the interactive Flash presentation.
TRUFORM 2.0 offers more flexible and powerful tessellation options than the original TRUFORM. Rather than being limited to fixed tessellation levels (1,2,3, etc.), TRUFORM 2.0 supports continuous tessellation, which allows floating point fractional tessellation levels for smoother transitions. An adaptive tessellation option is also supported, which dynamically adjusts the tessellation level of a surface depending on the distance from the viewer. Thus, nearby surfaces will have more polygons and more detail, while distant surfaces will have less.
TRUFORM 2.0 also supports displacement mapping, a technique that can be used to provide more control over the shape of 3D objects and surfaces. It works by modifying the positions of vertices according to values sampled from a special type of texture called a displacement map. The visual effect is similar to bump mapping, but much more realistic and detailed.
One of the most exciting features we saw with the Matrox Parhelia was the addition of the ability to take advantage of hardware displacement mapping that is native in DX9. While this may be a big deal to Matrox, I don’t think we are going to see a great amount of game developers take advantage of this technology outside using it to tessellate terrains in flight sims and such. I think hardware displacement mapping is liable to go the way of head casting and environmental bump mapping overall in that they are useful and flashy technologies that will never really gain wide acceptance and use. Still, while game devs may not fully embrace the technology from the software side, ATi has seemingly found a way to implement it from a hardware perspective. The original TRUFORM was a long ways from being perfect, but it was also a great step forward in delivering more lifelike characters. TRUFORM is something that must be programmed by the game developer, though. It is not something that is backwards compatible with older games and must be taken into account when coding newer games. We have not talked to game developers on this, but ATi does say that implementation is fairly simple. One thing that was confirmed is that TRUFORM will not be used in DOOM]|[ due to the shadows casted by characters not showing the TRUFORM model's visual outline. We still have not seen this feature in action nor did I see a demo showing it off. Seeing that it was really a feature that was not really talked about much, I doubt it is going to be something that is actually utilized much by the game developers or Radeon 9700 owners.
The 9700’s ability to use adaptive tessellation is exciting as well. Basically this would allow ATI to fully control the Level of Detail, or LOD, that we see as a character or entity about moves closer or further from the camera we are viewing it through. This ability by definition should allow for TRUFORM2 to be less of a load on the VPU.