No speedup with tensorrt 5 for fp32 performance vs. tensorflow

I have a model trained in tensorflow and see the same inference speed for FP32 in tensorflow and after TRT conversion. How can I go about debugging why there is no speedup?

I should say this is actually v5.1