Ati-Nvidia Difference on Normal Mapping

it happend this strange thing (maybe it’s not strange): in my program i pass to a normal mapping shader my camera and light position multiplied for inverse model matrix (position,rotation,scaling), then in my shader i trasform that to tangent space. On Ati card all function correctly, on Nvidia illumination is wrong, to make it function properly i have to multiply only light position for inverse matrix, is this functioning correct?