Installing CUDA 10.2 on Xavier NX

I was trying to install CUDA 10.2 on my Jetson Xavier NX but in the Nvidia download website there is no way for the arm64 architechture. I installed CUDA 11.0 but some issue was there for installing tensorflow libraries as I have Jetpack 4.4.1 and which doesn’t support CUDA 11.0. So is there any way to cross compile and install CUDA 10.2 on Jetson Xavier NX or building it from source code.

Hi jp,

You’ll need to use the binaries packaged with the JetPack you flashed the board with. You can follow this guide to do it.
https://developer.ridgerun.com/wiki/index.php/Installing_CUDA-toolkit_and_VisionWorks_Jetson_TX2/Xavier/Nano

Do take note that there’ll be generally a dependence between a given JetPack and the installed CUDA, so you’ll need to look for the TensorFlow binaries matching the CUDA version.

I have downloaded the below sources to build the Jetpack.
1.https://developer.nvidia.com/embedded/L4T/r32_Release_v4.3/t186ref_release_aarch64/Tegra186_Linux_R32.4.3_aarch64.tbz2
2.https://developer.nvidia.com/embedded/L4T/r32_Release_v4.3/t186ref_release_aarch64/Tegra_Linux_Sample-Root-Filesystem_R32.4.3_aarch64.tbz2

Unlike the prebuilt jetpack, this one does not come with CUDA pre-installed.

Is there any way to add the CUDA binaries to the jetpack while building the jetpack from sources?

Is there any solution for the aforementioned issue?