Originally Posted by moriz

GPUs are dramatically different from CPUs. one is a general computational device, the other is a specialized chip designed to accelerate the rendering of 3D objects on a 2D screen. you cannot compare the two based on clock speed, since they are so radically different.

the modern GPU is made up of a number of scalable graphics "cores", each can be programmed to execute generic code. this is a step up from the older, fixed function "shaders" design. this means, while GPUs are still primarily about accelerating graphics, they can also be used to do whatever you want them to. right now, the fastest graphic card on the market, AMD's radeon HD5870, has a whooping 1600 graphics cores. compare that, with the modern CPU, which is more or less limited to 4 cores. this means that graphic cards are the best processors for running parallelized code. if you can divide up your task to take advantage of this parallel processing power, a graphic card will do it dramatically faster than a CPU. in fact, a $5000 desktop computer, with four nvidia graphic cards in parallel, can outperform a multi-million dollar computing cluster by a few orders of magnitude on certain tasks.

video rendering/encoding/decoding are tasks that can easily be divided up into little chunks, which means that graphic cards, if properly utilized, are dramatically faster than CPUs in these tasks.