NVIDIA GeForce3 Investigation: What NVIDIA didn't tell us

Final Words

The vertex shader is a welcome breath of fresh air. One obvious future trend is extended instruction sets or increased instruction length as higher graphics processor performance becomes possible. Feedback of modified vertices may be important if precise collision or silhouette detection are required.

That brings us to the next issue. The shifting of geometry processing from the central processing unit to the graphics processing unit has several implications with regards to memory bandwidth. First, geometry traffic will further increase across the Accelerated Graphics Port bus (AGP) as the use of vertex shader becomes firmly entrenched. Because the graphics processor is good at what it does and as it shoulders greater responsibility, the demands placed on pixel bandwidth will become even heavier. To a certain extent, the pixel bandwidth load has been lightened by GeForce3's depth buffer optimizations but that still leaves much to be desired with regards to high resolution rendering and anti-aliasing with a full spectrum of colors.

The questions are: To what extent can current rendering architecture be leveraged? Is it possible to further optimize the mysterious Z-occlusion culling? What about color buffer compression? Perhaps a paradigm shift (buzzword alert!) is in order. Redefine the graphics pipeline on graphics application interfaces (API) with emphasis on early hidden surface removal and continue active developer support. NVIDIA would have the resources to do so, one suspects.

Special thanks goes out to Giovanni Caturano (Zetha), Markus Maki (Madonion) and Patric Ojala (Madonion) for their valuable inputs and insights.