Little performance difference between int8 and fp16 on RTX2080

[spolisetty]

Hi @fcj,

Hope this will help you.

YoloV4 slower in INT8 than FP16

Description Building my custom YoloV4 608x608 model in INT8 us slower than in FP16 on both my xavier nx and also on my 2080Ti. For example on the 2080Ti I get: FP16: 13ms per frame INT8: 19ms per frame Varying aspects of the INT8 calibration etc makes no difference to the speed. Is this normal ? Environment TensorRT Version: 7.2.1.6 GPU Type: RTX 2080Ti Nvidia Driver Version: 440.82 CUDA Version: 10.2 CUDNN Version: 8.2.0.53 Operating System + Version: OpenSUSE Python Version (if …

Could you please share complete verbose logs for both and if possible ONNX model to try from our end for better assistance.

Thank you.