AGX Xavier Cuda Installation

Robotics & Edge Computing Jetson & Embedded Systems Jetson AGX Xavier
1

Hi, I flashed my AGX Xavier without installing Cuda, there is no Cuda folder in usr/local/ directory. Now I want to install it, how can I do that, which version is compatible with SDK 5.1.3, is there any installation guide. My system is
aarch64
DISTRIB_ID=Ubuntu
DISTRIB_RELEASE=20.04
DISTRIB_CODENAME=focal
DISTRIB_DESCRIPTION=“Ubuntu 20.04.6 LTS”.

Note: When I tried to install it from SDK using JetPack 5.1.3 it says default ip is not available using USB cable, however by using ethernet, I can ping it.

Also How can I check if the Nvidia GPU drivers installed correctly. IN the Xorg.0.log it says:
[ 31.309] (II) NVIDIA(0): NVIDIA GPU Xavier (nvgpu) (GV11B) at SoC (GPU-0)
[ 31.309] (–) NVIDIA(0): Memory: 27603720 kBytes
[ 31.309] (–) NVIDIA(0): VideoBIOS:
[ 31.310] (–) NVIDIA(GPU-0): HP ZR24w (DFP-0): connected
[ 31.310] (–) NVIDIA(GPU-0): HP ZR24w (DFP-0): External TMDS

lspci | grep -i nvidia shows nothing

3

Hi,

You can install CUDA with apt command directly.

$ sudo apt-get update
$ sudo apt-get install nvidia-cuda

Thanks.

4

I followed your steps and now I have cuda 11.4 in the usr/local directory but nvcc --version command still not working.

5

also torch 2.0.1 couldn’t find cuda.

Python 3.11.10 (main, Sep 7 2024, 18:35:42) [GCC 9.4.0] on linux
Type “help”, “copyright”, “credits” or “license” for more information.

import torch
print(torch.cuda.is_available())
False

6

Hi,

Could you try to export the global environment and test nvcc again?

export PATH=/usr/local/cuda-11.4/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-11.4/lib64:$LD_LIBRARY_PATH

For PyTorch, you will need to install the package built with CUDA to use GPU.
But unforutenaly, we don’t have a package for Python 3.11.
We only provide the Python 3.8 package for JetPack 5 users.

Thanks.

8

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.