LOSTCIRCUITS

More features

The ASUS V6600 comes in two different flavors, the standard edition is basically the reference GeForce but the Deluxe edition has been juiced up with Video-In, two separate TV-Out (SVHS and composite RCA) and the ASUS VR100G 3Dglasses.

One specialty of the ASUS drivers comprises an idea adapted from CPU cooling programs: "ASUS SmartCooling". This utility is a diagnostic tool measuring voltage and temperature on the GPU and, in the event that a critical margin is exceeded, reduces the operating frequency of GPU and memory on the fly until normal values are back. The claim is that this utility allows safer overclocking since the power will be used only when needed. In principle a good idea but the utility itself usually monitors the state of things only at startup. In order to provide continuous monitoring it needs to be set as Resident which allows to define the monitoring intervals from a few seconds to minutes. The only drawback is that SmartCooling itself causes a performance hit that counteracts the performance gains of overclocking.

There has been a substantial amount of writing on this subject and we don’t want to regurgitate what has been said before. It is necessary, though, to at least give a brief description of what the GPU stands for. GPU or graphics processing unit is another word for the old-fashioned graphics processor or rather, it is not. There are some major differences between the conventional graphics processor that we know and the GPU. In a few words, conventional graphics processors process the triangle setup (the 3-dimensional representation of the world) and the clipping (the cutting of polygons at the edge of the screen) as well as the actual rendering of the image. Transformation and lighting of the scene are still a task for the CPU.

To briefly summarize the entire process, every representation of a virtual world on the screen usually involves 6 steps

• Application tasks (the movement of objects according to tasks, movement of camera, aim of camera)
• Scene level calculations (selection of detail level, object level culling, creating object mesh)
• Transform
• Lighting
• Triangle setup and clipping
• Rendering

The first two steps are necessarily performed by the CPU. That is, the graphics processor has relatively little saying in whether one decides to move forward in a 3D environment or whether the level of detail in the background is high or low, depending on the distance.

The next two steps, transform and lighting (or T&L) have, by convention, also been dedicated tasks of the CPU but here is where the GPU changes a lot, if not everything.

Transforming a given scene into a visible image is an extremely calculation intense application because every surface is composed of polygons. The quality of the tesselation (the process of breaking up a curve into small line segments) depends on the number of segments generated. The performance, on the other hand, is inversely correlated to the number of segments contributing to the facet. This is basically just saying what everyone knows, that better quality causes a performance hit. Add specular lighting, taking into account the direction of the light source as well as the position of the viewer and there is an enormous amount of data that need to be calculated.

The speed of CPUs historically doubles in an eighteen months cycle. The speed of graphics engines, within the same time frame, almost quadruples. Because of the lack of CPU power, professional graphics workstations have started to uncouple T&L from the application tasks by means of dedicated, specialized processors. These separate T&L setups showed that a garden variety CPU is particularly unsuited for this task. Regardless of the extension set in form of SSE, MMX or 3DNow!, CPUs are too multipurpose oriented than to make good use of their power for T&L. Instead, a much less potent but highly specialized graphics processing unit can outperform any given CPU in its own terrain. The beauty of it is that, whoever makes it, can also implement the necessary drivers for the instruction set without being at the mercy of third party manufacturers (sounds familiar?).

CPU power has increased tremendously over the past few years but compared to photo-quality, computer graphics are still in the stone ages. At the same time, a critical mass has been reached in the performance vs. quality conflict. More precisely, most PCs nowadays are fast enough to deliver the standard 30-60 fps in any given game and the fill rates of the graphics adapters are good enough to at least double these frame rates. In other words, it is time to move on to the next level of visual quality and the easiest way to achieve that is the GPU with its own T&L engine.

How is all of this accomplished?

Here is a clip from ASUS webpage:

The 256-bit graphics architecture is equivalent to four 64-bit rendering pipelines that is capable of delivering four pixels out per clock. This is the major hint for the chip name GeForce"256". Each 64-bit data is comprised of 32 bits for color (8-bit each for red, green, blue, and alpha), and another 32 bits for both Z and stencil buffer (all 32 bits for Z, or 24 bits for Z and 8 bits for stencil). With the 32-bit Z buffer, some annoying artifacts previously caused from the insufficient Z resolution can be easily resolved. The DRAM controller of GeForce256 has the capability for handling the memory size up to 128MB. The possible DRAM sizes to be mounted on a GeForce256 based graphics card are 8, 16, 32, 64, and 128MB. Furthermore, the 350MHz RAMDAC is good for accomodating with better monitors to achieve higher resolutions and refresh rates

A few more features
(just for completeness and because they are simple solutions for complicated tasks.)

Cube environment mapping \ sphere environment mapping

A 3D space can be represented in form of a sphere, like someone walking in circles around an object. However, 3D means that there are three dimensions and that means that every point can be represented by using 6 possible directions (left / right, up / down, forward / backward) which is much easier to calculate than a spherical space and results in less distortions. Cube environment mapping is an integral feature of DX7 as are vertex blending and particle systems (as the name says, if you blow things into particles).

All in all, the concept sounds like a step into the right direction and there are a few predictions that can be made about the performance, based on the above said.

• 1 the simpler the game and the graphics, the less will be the performance gain.
• 2 in complex applications, the gain will be much greater
• 3 applications that specifically address GPU and the separate T&L engine will show an enormous performance boost.
• 4 the bus transfer of data from application and scene level tasks will play a more important role in the overall performance of the system.

Keep these speculations in mind when you read the benchmark results.

Next Page:    => installation =>