Both, actually... both tend to be fairly memory hungry if the polygon count is high, but in both cases, the critical thing is physical memory. If your machine doesn't have enough physical memory, you could easily make even a dual 1.5 GHz Itanium2 perform like a handheld. Performance-wise, the main difference between the modeling and rendering processes is that the modeling process requires sending a lot of data to the graphics card, while the rendering process, at least when using a raytracer, is generally limited by computing power first. In either case however, virtual memory access is slow enough to be the dominant factor if there isn't enough physical memory to cache the entire scene.