Jetpack3.1 install TensorRT2.1 take place of TensorRT1.0 error

Hello, I have a question, I have a TX1, used Jetpack3.0 flash, but because I want to use TensorRT2.1 now, so I only through Jetpack3.1 TensorRT2.1 installation, but cannot run samples , and I can’t find my video card driver


Please also reflash rel-28.1 OS instead of only installing TensorRT package.

Does this mean that software in different JetPack packages can’t coexist, and do I need to start from scratch?


We don’t officially support this kind of combination.
Not sure if there is an issue or bug. Maybe other users can share some experience on this.

We recommend user to install OS and package from the same JetPack release to make sure all things work correctly.

Thank you for your reply, I now use Jetpack3.1 after brushing, test caffe - yolo project, compile successfully, but haven’t started to read the images in the process of testing process is killed,test darknet - yolo program, which I used to run through the code on TX1, killed too

I wonder if there is something wrong with the flash

If there is a message about killing a program, then you can expect that by far the highest likely cause is running out of memory and having the system kill it. Run something like “htop” while doing this and see if memory usage goes up.

I have run “top”to view of memory, also killed when the remaining 500 M.But if I only use CPU compilation, operation can run, but the results is wrong, but flash before this project can be run successfully, if I broke a part TX1 when brushing

I would like to ask, is it because when I use jetpack3.1, the host system USES Ubuntu16.04?


Please remember to use the packages from the same JetPack version.

TensorRT 1.0 is built with CUDA-8.0.34/CUDA-8.0.64 while JetPack3.1 uses CUDA-8.0.84.
Different CUDA version may lead to driver error.

Please use the package from the same JetPack version to ensure the functionality.