Carmack Technical Questions

Beyond3D has some extremely detailed questions with John Carmack in John Carmack on DOOM3 rendering. Of interest is the mention that “[t]he game characters are between 2000 and 6000 polygons”, which makes sense, given the massive number of rendering passes in Doom 3. Since all shading and lighting is done in real-time, the number of polygons becomes a lot more important.

Other engines, like the Unreal engine, boast much higher character-polygon counts, but they use only limited dynamic lighting. Of course, they generate nice shadows for the models, but the shadows won’t interact nearly as well with the rest of the environment and the other models as the Doom engine promises to do.

With the heavy use of bump-mapping to generate lifelike, photorealistic detail with extra polygons, Doom’s use of dynamic lighting will still produce much more lifelike scenes, despite only a small increase in number of polygons per model. With the new, fully real-time rendering system, more video horsepower will always help. He mentions that “[t]otal number of render passes is (greatly) influenced by the number of per pixel lights used.” and, since the engine’s render path is far more straightforward that it used to be, the expected speed of a scene is somewhat easily predictable. The speed will be proportional to the number of lights used, excepting some unusual cases, like “making very jagged models with lots of little polygonal points”, which would slow things down inordinately as the engine rendered a lot of little shadows on a lot of little surfaces.