tensorrt=10.13.2.6-1+cuda12.9 in nvidia repo cannot be installed after the release of CUDA 13.

[xkszltl]

Description

This may applied to other version of trt as well.

The following command failed to execute today (Aug 2025) after the release of CUDA 13.
Looks like the solver always picks up +cuda13.0 first and fail.
tensorrt should explicitly specify its supported dependencies on libnvinfer/libnvonnxparsers or add conflict to higher major version.

# sudo DEBIAN_FRONTEND=noninteractive apt-get -o DPkg::Lock::Timeout=3600 install --allow-downgrades tensorrt=10.13.2.6-1+cuda12.9
Reading package lists... Done
Building dependency tree... Done
Reading state information... Done
Some packages could not be installed. This may mean that you have
requested an impossible situation or if you are using the unstable
distribution that some required packages have not yet been created
or been moved out of Incoming.
The following information may help to resolve the situation:

The following packages have unmet dependencies:
 tensorrt : Depends: libnvinfer10 (= 10.13.2.6-1+cuda12.9) but 10.13.2.6-1+cuda13.0 is to be installed
            Depends: libnvinfer-plugin10 (= 10.13.2.6-1+cuda12.9) but 10.13.2.6-1+cuda13.0 is to be installed
            Depends: libnvinfer-vc-plugin10 (= 10.13.2.6-1+cuda12.9) but 10.13.2.6-1+cuda13.0 is to be installed
            Depends: libnvinfer-lean10 (= 10.13.2.6-1+cuda12.9) but 10.13.2.6-1+cuda13.0 is to be installed
            Depends: libnvinfer-dispatch10 (= 10.13.2.6-1+cuda12.9) but 10.13.2.6-1+cuda13.0 is to be installed
            Depends: libnvonnxparsers10 (= 10.13.2.6-1+cuda12.9) but 10.13.2.6-1+cuda13.0 is to be installed
            Depends: libnvinfer-bin (= 10.13.2.6-1+cuda12.9) but 10.13.2.6-1+cuda13.0 is to be installed
            Depends: libnvinfer-dev (= 10.13.2.6-1+cuda12.9) but 10.13.2.6-1+cuda13.0 is to be installed
            Depends: libnvinfer-lean-dev (= 10.13.2.6-1+cuda12.9) but 10.13.2.6-1+cuda13.0 is to be installed
            Depends: libnvinfer-plugin-dev (= 10.13.2.6-1+cuda12.9) but 10.13.2.6-1+cuda13.0 is to be installed
            Depends: libnvinfer-vc-plugin-dev (= 10.13.2.6-1+cuda12.9) but 10.13.2.6-1+cuda13.0 is to be installed
            Depends: libnvinfer-dispatch-dev (= 10.13.2.6-1+cuda12.9) but 10.13.2.6-1+cuda13.0 is to be installed
            Depends: libnvonnxparsers-dev (= 10.13.2.6-1+cuda12.9) but 10.13.2.6-1+cuda13.0 is to be installed
            Depends: libnvinfer-samples (= 10.13.2.6-1+cuda12.9) but 10.13.2.6-1+cuda13.0 is to be installed
            Depends: python3-libnvinfer-dev (= 10.13.2.6-1+cuda12.9) but 10.13.2.6-1+cuda13.0 is to be installed
            Depends: libnvinfer-win-builder-resource10 (= 10.13.2.6-1+cuda12.9) but 10.13.2.6-1+cuda13.0 is to be installed
            Depends: libnvinfer-headers-python-plugin-dev (= 10.13.2.6-1+cuda12.9) but 10.13.2.6-1+cuda13.0 is to be installed
E: Unable to correct problems, you have held broken packages.

Workaround is to manually specify dependencies aside from the meta package:

apt-get install {{,python3-}libnvinfer,libnvonnxparsers,tensorrt}'*=10.13.2.6-1+cuda12.9'

Environment

Operating System: Ubuntu 22.04

[kevinch-nv]

This is a known limitation of the apt install method of TensorRT. I see you already found the work around to install the CUDA 12.9 packages, but https://docs.nvidia.com/deeplearning/tensorrt/latest/installing-tensorrt/installing.html#using-the-nvidia-cuda-network-repo-for-debian-installation has some more information

[xkszltl]

How about adding version constraints to the dependency in deb?