Picture this, you are buying a GPU that can peddle a bicycle and there are different speeds that different riders can go based on their power. Now picture your CPU as the bike itself. Even the faster rider won't be able to go faster if they are on a bike made for a toddler. Put that same rider on a bigger bike and they go faster. Now put a less fast rider on the same two bikes. The slower bike rider may go just as fast as the faster rider on the smaller bike but once they are put on the adult sized bikes the faster, stronger rider will clearly pull away. Hence if the rider and his bike are not evenly matched the output will be effected substantially. Now also picture this, there are many different adult bikes as well but there are Wal-Mart bikes and pro bikes that are 100x the cost. However there is a point at which the bike wont make the rider faster, it will then rely solely on the power of the rider him/herself. This is the easiest way to think about computer bottlenecks. The GPU will want to go 100% of its capacity but might be limited by the CPU it has to go "through", just as the rider can peddle and peddle but only goes as fast as his bike physically allows. In gaming you want to allow your GPU to perform without any hindrance, aka a bottleneck, and for each GPU there will be its own certain threshold. I used to have the GTX 650 ti Boost and paird it with both a i3-2120 and a i7-3770k. Neither of those were bottlenecks btw. Also the i7 didn't change the FPS whatsoever. If I had gone down to a Pentium, who knows, maybe it would have been bottled but maybe not. Hope this analogy helped.