As stated in the post
I’m planning to deploy TensorRT7 as a dll plugin for a portable application on Windows platform, and I found out by looking at the VS calls that TensorRT actually calls a lot of dll files from the CUDA\bin directory, and I think I can only install CUDA to run the TensorRT portable application.
My question is, is it possible to satisfy a portable application with a minimum of dll dependencies? Not by installing CUDA
Environment
TensorRT Version: 7.2.2.3 GPU Type: RTX 3080 Nvidia Driver Version: 470.05 CUDA Version: 11.1 CUDNN Version: 8.0 Operating System + Version: Windows 10 21343
Thank for your reply
We can install and run TensorRT for Windows platform on our model properly, but we are aiming to deploy on Windows platform and need to consider the case where CUDA is not installed.
However, I tried copying the dll from the CUDA\bin directory that the project needs to the project’s lib folder, then deleting CUDA from Path and adding only the project’s lib directory to Path, which does run and does not require CUDA to be installed on Windows
I’m also interested in the list of dependent dlls for easy deployment on Windows platforms. I could not find all dependencies with Dependency Walker as it freezes on my system. I had to determine the needed dlls through experimentation. The dlls that I was able to find are:
cublas64_11.dll
cublaslt64_11.dll
cudart64_110.dll
cudnn64_8.dll
cudnn_cnn_infer64_8.dll
cudnn_ops_infer64_8.dll
nvinfer.dll
nvinfer_plugin.dll
nvonnxparser.dll
nvrtc-builtins64_111.dll
nvrtc64_111_0.dll