Can we install Tensorrt on jetson AGX orin without reflashing the GPU?

[tonysalloom]

My system doesn’t have tensorrt installed, I’m trying to reinstall it without reflashing the GPU, is that possible? and how to do it?

[AastaLLL]

Hi,

Could you try to run the below command to verify the TensorRT installation first?
If the tool can run correctly, TensorRT should be installed on your environment already.

$ /usr/src/tensorrt/bin/trtexec --help

Thanks.

[tonysalloom]

Thanks for your reply
I don’t have tensorrt folder in /usr/src/ ,it’s been removed somehow.
the output of your command is “no such file or directory”.
is there anyway to install it without reflashing the gpu

[tonysalloom]

I’ve downloaded the tar file for tensorrt 8.5.3.1 for Ubuntu-20.04. aarch64, cuda 11.8. cudnn8.6, unzipped it and copied the folder to the path /usr/src/.
Next, I did export paths as follows:

$ export LD_LIBRARY_PATH=/usr/src/tensorrt/lib:$LD_LIBRARY_PATH
$ export  PATH=/usr/src/tensorrt/bin:$PATH
$ export  /usr/src/tensorrt/include

Now the command ```
/usr/src/tensorrt/bin/trtexec --help

&&&& RUNNING TensorRT.trtexec [TensorRT v8503] # /usr/src/tensorrt/bin/trtexec --help
=== Model Options ===
  --uff=<file>                UFF model
  --onnx=<file>               ONNX model
  --model=<file>              Caffe model (default = no model, random weights used)
  --deploy=<file>             Caffe prototxt file
.... 
etc.
I'm not listing all contents to save the space.
Now when trying to build my project, It fails with the following messages:
[10/25/2024-09:25:21] [W] [TRT] Unable to determine GPU memory usage
[10/25/2024-09:25:22] [W] [TRT] Unable to determine GPU memory usage
[10/25/2024-09:25:22] [I] [TRT] [MemUsageChange] Init CUDA: CPU +8, GPU +0, now: CPU 20, GPU 0 (MiB)
[10/25/2024-09:25:22] [W] [TRT] CUDA initialization failure with error: 222. Please check your CUDA installation:  http://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html
[10/25/2024-09:25:22] [E] Builder creation failed
[10/25/2024-09:25:22] [E] Failed to create engine from model or file.
[10/25/2024-09:25:22] [E] Engine set up failed
&&&& FAILED TensorRT.trtexec [TensorRT v8503] # trtexec --onnx=model/resnet50int8/camera.backbone.onnx --fp16 --int8 --inputIOFormats=fp16:chw,fp16:chw --outputIOFormats=fp16:chw,fp16:chw --saveEngine=model/resnet50int8/build/camera.backbone.plan --memPoolSize=workspace:2048 --verbose --dumpLayerInfo --dumpProfile --separateProfileRun --profilingVerbosity=detailed --exportLayerInfo=model/resnet50int8/build/camera.backbone.json

[AastaLLL]

Hi,

We don’t release Jetson-compatible packages for the JetPack 5 users on the website.
Suppose you download some SBSA package that cannot work on Jetson as it is for dGPU while Jetson is an iGPU system.

Please remove the package and install it via the apt command instead.

$ sudo apt update
$ sudo apt install nvidia-tensorrt-dev

Thanks.

[tonysalloom]

I did reinstalled the jetpack. Now I have the following :

Where can I find the lib path and the include folder of tensorrt ?

[tonysalloom]

Thank you, problem solved.