Why tensorRT 8.2.1-1+cuda10.2 does not exist for arm64?

[bayab]

Description

Hello,

I wanted to know why Jetpack 4.6.2. comes with CUDA 10.2 and TensorRT 8.2.1 but there is actually no possibility to download TensorRT 8.2.1 with CUDA 10.2 from the TensorRT download page for
arm64 architecture: https://developer.nvidia.com/nvidia-tensorrt-8x-download.

This causes all sorts of issues when trying to install TensorRT 8.2.1 at the container level using a Nvidia Cuda 10.2 base image.

Thanks!

[AakankshaS]

Hi @bayab ,
You can use the below link to download trt 8.2 with CUDA 10.2
http://cuda-repo/release-candidates/Libraries/TensorRT/v8.2/

Thanks

[bayab]

Hi AakankshaS,

It seems the link you provided is broken?

[AakankshaS]

Hi @bayab ,
Updated response -

@bayab

Jetpack 4.6.2. comes with CUDA 10.2 and TensorRT 8.2.1 but there is actually no possibility to download TensorRT 8.2.1 with CUDA 10.2 from the TensorRT download page for
arm64 architecture: https://developer.nvidia.com/nvidia-tensorrt-8x-download

aarch64 packages are only available through the Jetson SDK manager, hence cant be downloaded externally.

Thanks

[bayab]

Thanks for your reply. Actually for other people trying to accomplish this, one way to do it is to flash your Jetson with Jetpack 4.6.2 and then manually copy all required nvidia libraries into a container then save that image, in this way you do not need to download tensorRT 8.2.1 and cuda 10.2 from external sources in your Dockerfile.