Most of my graphics works was better served by an understanding of hardware(pipelining, SRAM vs DRAM caches and latency penalties including random vs linear reads). Most of the embedded work was done with a rigor towards memory management and not any "traditional CS". In fact we avoided traditional CS structures(lists, trees, etc) since they had negative performance penalties. The only thing that we came close to leveraging was radix sort along the view axis, and that was through a gnarly hack that let us use it for floating point values.

I've also yet to run into anyone in gamedev or the graphics world who could derive quaternions from first principals. They're damn useful but I think you you know their operations understanding the theory doesn't take you much further.

I think you have a point on rigor, but that has more to do with how to tackle problems broadly rather than any sort of academic or theoretical background.