nVidia release versions compatibility


I’m struggling with nVidia releases. Lets say, I want our product to use TensorRT 8.5.2, deploying in an official nVidia TensorRT container.

  • The available TensorRT downloads only support CUDA 11.8
  • The v23.01 of the container, the first version to support 8.5.2 (v22.12 is 8.5.1), ships with CUDA 12.0.1.

What is the expectation here? Mine are that either the development package is compatible with the Docker image, or vice versa. Too much?


TensorRT Version:
GPU Type:
Nvidia Driver Version:
CUDA Version:
CUDNN Version:
Operating System + Version:
Python Version (if applicable):
TensorFlow Version (if applicable):
PyTorch Version (if applicable):
Baremetal or Container (if container which image + tag):

Relevant Files

Please attach or include links to any models, data, files, or scripts necessary to reproduce your issue. (Github repo, Google Drive, Dropbox, etc.)

Steps To Reproduce

Please include:

  • Exact steps/commands to build your repro
  • Exact steps/commands to run your repro
  • Full traceback of errors encountered

Hi ,
We recommend you to check the supported features from the below link.

You can refer below link for all the supported operators list.
For unsupported operators, you need to create a custom plugin to support the operation


Sorry, but I don’t believe this answers my question. My question was about 3-way release compatibility between TensorRT, CUDA and TensorRT Docker image, specifically when applied to v8.5.2 of TensorRT.

Your answer is about ONNX operations compatibility in TensorRT 8.6.

Did I miss some part of it which answers my question?

Phrased differently: if I use TensorRT 8.5.2 in development, which version of nvcr.io/nvidia/tensorrt should the resulting software be deployed on – considering v22.12 of it still uses TensorRT 8.5.1, and v23.01 of it already wants CUDA 12.0.


We recommend that you please use the latest TensorRT version 8.6.1.

Please refer to the notes added in the support matrix document.

If we have supported driver version, we can use the CUDA versions accordingly.