I already know of utilities like
htop that can be used to (roughly) tell if a process is primarily bottlenecking at disk read/write speeds, network IO, not being parallel, or being parallel but running out of cores nevertheless.
However, is there a way to do something similar for the graphics card? Part of the problem, admittedly, is that if the GPU is actually being used for the display, any way to view status might affect the GPU use by hiding whatever program was rendering something complex to the screen. But even outside of that case*, I haven't been able to find any such tool. Perhaps I've just been looking incorrectly. Does such a tool exist?
*(GPU used for non-graphics math, or a small animated-graph-icon for the status display that doesn't hide anything significant or add much overhead)