Questions about the performance of test cases in test_trt_ops.sh

[nainaigetuide]

When deploying on an X86 device using the official image and testing test_trt_ops.sh, it is found that the time for int8 precision is not lower than that for fp16 precision. Why is that? @DerryHub @firestonelib