X-bit labs

NVIDIA GeForce FX 5950 Ultra against ATI RADEON 9800 XT: Shader Wars (page 4)

Category: Video

by Tim Tscheblockov

[ 10/23/2003 | 03:31 PM ]

NVIDIA Drivers: We Know to Optimize Shaders

As is known, the DirectX 9.0 Pixel Shader performance of NVIDIA GeForce FX graphics processors is not as high as that of ATI RADEON 9500/9700/9800 chips. There are several reasons for that. First of all, lower Pixel Shader performance is the price you have to pay for higher flexibility of GeForce FX architecture (see our NVIDIA GeForce FX 5800 Ultra Review: New Technologies and Performance and NVIDIA GeForce FX 5900 Ultra (NV35) Review for more details on that). Secondly, it is no secret that DirectX 9 specifications were developed by Microsoft, which cooperated closely with ATI, shortly before the R300 chips series announcement. Therefore, it is not at all surprising that the DirectX 9 shaders are often “easier” for ATI graphics processors than for NVIDIA chips with a completely different architecture. As a result, it turns out that the shaders with the same input and output parameters and performed in the most optimal way on both: GeForce FX and RADEON 9X00, look differently.

In order to increase the Pixel Shader efficiency, NVIDIA developed an optimizing shader compiler, which takes into account all highs and lows of GeForce FX architecture and optimizes the process accordingly.

This is what the way the algorithm for “DirectX 9 Pixel Shader – GPU” looks like according to NVIDIA:

DirectX 9 Pixel Shaders are not used directly by the graphics processor. At first, DirectX translates the shader into a universal code and only then the graphics card driver compiles the code into pixel processor commands.

Exactly during the compilation of this intermediate code the new NVIDIA’s technology starts working. I am talking about NVIDIA Unified Compiler Technology, which is included into the driver called ForceWare. The optimizing compiler should take into account all GeForce FX peculiarities, so that it could create a chain of commands to be performance by the GPU as fast as possible.

For this purpose the compiler “takes apart”, analyzes and then “puts back together” the shader and can even rearrange the commands into pairs of texturing and mathematical tasks, reduce the number of involved time registers, reduce the overall number of commands by using only those of them, which are supported only by the GeForce FX hardware, etc.

The optimization criterion is the maximum processing speed of the output code. Besides, the output parameters of the initial and optimized shader should be in perfect coincidence. In other words, NVIDIA claims that the use of optimizing compiler eliminates any possibility of quality worsening.

This approach deserves all the applause. Firstly, the optimizing compiler can work with any shaders without any exceptions, and not only with those that are used in 3DMark093, for instance. Secondly, the company promises that the image quality will not suffer at all.

Well, when we compare the performance of NVIDIA GeForce FX 5950 Ultra with ATI RADEON 9800 XT, we will see if NVIDIA’s software developers managed to achieve all these goals.