Yes that paper discuss CPU AND GPU bottlenecks. And how you decrease bottlenecks. But where do you see that that one bottleneck stands for 100% of the total time? If you have a video card running at 2000 GHz it will maybe render the picture in notime. Almost all time that needs to render the picture will be the processor time. But even if the GPU is extremely fast it will need to add a fraction of the total time.
If you have something that is slow you will of course look for areas that takes most time. But even if something takes very much time it doesn't use 100% of the time.
Bookmarks