Video encoding is not parallel enough to be worth encoding on the GPU itself. GPU encoders (implemented in CUDA etc. and running on the GPU itself) are a thing of the past, the tradeoffs they made to achieve the encoding speed people expected had a significant negative effect on compression efficiency (not to mention it would impact graphical performance of other apps while encoding).
Modern hardware-accelerated encoders may not be the most efficient out there (in terms of compression) but are still significantly better than the old generation.