GPU-like inference performance on CPUs in production for computer vision

Can you achieve GPU-like inference performance on CPUs in production for computer vision and natural language processing tasks by c++ and python ?

onnx runtime , torchscripts , openvino , ncnn , tensorart , ....