how to install tensorrt library independently?

Hi, I check xavier platform tensorrt version is 5.0.6. It is lower version, so I remove this version lib, Now, I want to install higher version of tensorrt version on Xavier, but do not want to flash it again. Can you support install tensorrt library on xavier independently? Thanks.

Hi wupeng, the TensorRT version has dependencies on cuDNN and CUDA version, which in turn has dependencies on the L4T driver version, so it isn’t possible to upgrade TensorRT without upgrading JetPack. It’s recommended to upgrade to the latest JetPack version to upgrade TensorRT.

Note that you will have to re-flash your device to upgrade to JetPack 4.3, but future updates to JetPack will be delivered through an APT package server, so in the future it will be possible to upgrade JetPack from the command line without needing to re-flash. However to get your device onto JetPack 4.3 you will need to re-flash, so remember to backup any code or files that you wish to save beforehand.

Hi, dusty_nv. Thanks for your answer. I re-flash my Xavier with JetPack 4.3, After the re-flash my device, this tensorrt version is 6.0.1. I input this command: dpkg -l|grep TensorRT is right result. But I using cmake to compile TensorRT, it errors. the errors messages as follows: Could Not find TensorRT: Found unusitable version “…”, can you known why this errors happened ?

Hi wupeng, can you post your CMakeLists.txt file? Or are you trying to compile a TensorRT sample?

On Jetson, the TensorRT headers are found in /usr/include/aarch64-linux-gnu and the TensorRT libraries are in /usr/lib/aarch64-linux-gnu (both of these paths should already be used by the compiler automatically)

Here is my CMakeLists.txt that links with TensorRT library, this is the only needed code for it:
https://github.com/dusty-nv/jetson-inference/blob/8846dcb274ff812880ef7bf9d42eb1dd75c9e7b7/CMakeLists.txt#L150

Hello, I was wondering if this is possible as well. I was hoping to install Tensorrt 6 on Jetson Xavier, without re-flashing Jetpack. Is there a way to do this? Thank you very much.

Hi matthew_berg,

No, all those SDKs can only be installed via JetPack.