TensorRT inference is slower under multi-threading

Hello, what effect does thread switching on the host have on tensorrt inference? I’ve found that on my server, using a multi-threaded inference model takes longer than a single-threaded inference time, which doesn’t include c’u’d’aCopy time, and G’P’U is less utilized;

Hi @2235636388 ,
Apologies for the delay, I am checking on this, and shall revert soon.

Thank you