On a GPU you often have 500+ cores, and using a pcie x16 bus, you wont have a bottleneck there.
The GPU is build for parallel processing, with a architecture that support the load, with a bandwith of +50GB/s
Add the fact that most new cards use DDR5 ram, rahter than DDR3 as you will find on motherboards
Summ it up, and you will see why GPU acceleration is the future 🙂
Very few software supports it well right now, but its a question of time, since the demand is growing