Transformation & Lighting

Triangles Everywhere!

Let's cut to the chase. Theoretically, today's high end processors can only push out 3-6 million triangles per second. Id Software's John Carmack stated in his 9/2 .plan update that Quake 3 targets about 10,000 triangles per frame. If we get a healthy 100 frames per second, that means the processor has to push out about 1 million fully lit and transformed triangles per second to maintain that 100fps. The trouble is that the CPU has other duties besides pumping out triangles. The CPU also has to run the game, calculate physics, handle the AI, etc... -all this in addition to keeping the OS and any other background programs happy.

Current generation 3D accelerators only have a triangle setup engine. We have to wait for the CPU to keep track of all the object entities in the 3D world, perform object culling (determine what objects are in view), transform the 3D coordinates into a 2D scene, and add lighting before the the graphics card finally gets into the act.

Relax, Take a Load Off!

Next generation cards such as the GeForce and Savage2000 move the transform and lighting steps onto the graphics card. All the CPU has to do is feed the GeForce the 3D coordinates of the triangles and the GPU handles the rest, transforming the 3D scene into a 2D frame and applying the lighting.

More Triangles = More Detail

The highly specialized GeForce GPU T&L engine can process about to 10-15 million triangles a second. Currently, game developers must watch the number of polygons they use because they don't want to hurt performance. Developers must walk a fine line between having detailed levels and models and having the game run at a playable framerate. The GeForce 256 GPU allows developers to use many more triangles to create incredibly detailed scenes.

Hardware T&L support is already built into OpenGL and DirectX 7, but it'll be a while before we start seeing games that will take advantage of all the triangles the GeForce is capable of handling.