We have a system based on legacy JetPack 3.2.1. The JetPack has GPU driver for Cuda 9 and TensorRT 3.
We want to use TensorRT 7.1.3 which requires CUDA 10.
I can install Cuda, CuDNN and TensorRT deb packages from JetPack 4.5.1 on TX2 with JetPack 3.2.1, however when running Cuda examples they complain about old version of Cuda driver (which is logical kernel has CUDA9 driver and we want CUDA10).
The question is would it be possible at all to upgrade CUDA driver in the kernel so it will allow to run CUDA10? Simple upgrading the JetPack is problematic due to the fact that we have custom kernel modules compiled for the old JetPack.