Ah, I thought I'd stop before going into too much detail, but the above is wrong re the CPU. For example, with Adobe products, you need a GPU that is supported by Adobe (most are), and you need to make sure the VRAM is comfortably larger than a frame plus lots of overheads that the Adobe site can tell you about. At that point, the software will offload video transforms to the GPU. I don't know about recent versions, but until about a year ago, if the GPU failed to cope with any frame, the load would be dropped back to the CPU for the rest of the job: it couldn't go back to the GPU..
When we're rendering, doing it in the GPU is about 50-200 times faster than the CPU: there's absolutely no comparison. But we're suddenly getting so far away from the original question.
Blender has something similar. ON NVidia you would look for CUDA support (everything from GTX 4xx onwards) and on AMD you would check for OpenCL support on GCN (which is HD, er, 7000 and above, with exceptions.)
But anyway, for the home user with home video, everything works, it may just take time. But for much video work, an extra $150 on a GPU does more than $150 on a CPU.