If you use a cube as emitter, what else did you expect than a grainy light? Remember the light on a location is only made from a limited number of samples. One pixel 'sees' 3 rays of light from the bottom of the cube and 1 from the side. The next 'sees' 3 samples from the side and only 1 from the bottom. Angles to normals of bottom and normal of side are significantly different so gives different amount of light. Use a sphere as emitter for more even result.