Fetch unit refers to the hardware unit responsible for retrieving data from the global buffer/VRAM (in this case, it is either a vertex fetch unit or texture fetch unit in the GPU).
The profiler calculates the values by taking the average of all the fetch unit cycles compared to the cycles that the GPU is active.
FetchUnitStalled counter is the average number of cycles fetch instructions are waiting for the fetch units to be available (the units might be unavailable due to waiting for data to return for other instructions) compared to the cycles that the GPU is active.
I hope that helps.
bpurnomo, Thank you.