CPU vs GPU neural network recognition?

Question

Neural networks

Greetings!
At the moment, recognition takes place on the CPU (12 cores) through Yolov5 and the load on it, to put it mildly, is not weak. The GPU is simple, only for graphics.
I did not find a specific answer in Google, mostly infa regarding training.
But as far as I know, the processor outperforms the video card in calculations.
I'm interested in the following question:
Does it make sense to go broke on a good video card with tensor cores?
Will it effectively affect the processing speed or should we not expect improvements?

The question is about recognition, not learning.