Light pre-pass
A deferred-lighting variant that stores only depth and normals, lights into a buffer, then re-renders geometry to apply material — a thinner G-buffer than full deferred shading.
Full deferred rendering stuffs the whole material — albedo, specular, normals — into a fat G-buffer and then shades. Light pre-pass instead splits lighting into two passes: first a thin pass that writes only depth and normals and accumulates all lighting into a buffer, then a second pass that re-renders the geometry and multiplies each material's albedo by the accumulated light.
Wolfgang Engel described the technique around 2008. It uses less bandwidth and fewer render targets than full deferred shading, at the cost of submitting the geometry twice. Once MRT bandwidth grew cheap, most of its advantage moved on to Forward+ and clustered shading.