Fool Running is right. Try to find some tutorials on Frustum Culling - that is the metod of displaying only visible polys of terrain. LOD method is another optimisation. You divide terrain into chunks (lets say 32x32 points), and for every chunk you pre-compile how many you want LOD parts - with 32x32, 16x16 and 8x8 points for example. Then, if your camera is far away from chunk (AND it is visible), display lower detail version of chunk.
More memory consumation - but better performance!