About acceleration with TensorRT5.1.5.0 on Tesla T4

Windows 10
cuda 10.0
cudnn 7.5
caffe yolov3
I created the int8 engine file for yolov3 detection and it cost about 90ms one image, that is 60ms on GTX 960.What is the reason?


Can you share your code/script and model file so I can try to reproduce this or further debug?