I assume since all voxels are the same size you can just batch draw them but even so there is going to be a lot within viewing distance that may be invisible.
Are they all just assigned to an octtree and then culled against the view frustum or do you really need some clever occlusion culling?
Minecraft for a long time just batched each visible side, I think even ignoring occlusion, and drew it; I don't know if they precompute larger meshes now.
[1] - http://static.largerussiangames.com/voxels/videos/rigidbodie... (Each 'companion cube' is really attached to a spherical physics body, the physics aren't mapped to the mesh.)
Also there was https://en.wikipedia.org/wiki/Comanche_series even before Outcast. I don't think it used polygons. It was launched years before 3D graphics cards, too..