How to install CUDA 10.2 for Aarch64 (used by Jetson) within l4t-base container?

Accelerated Computing NGC GPU Cloud Docker and NVIDIA Docker
, ,
1

I am running an NVIDIA Jetson Container (l4t-base:r.32.5.0) to develop software for NVIDIA Jetson. Since I am missing some CUDA libraries, I intended to install the nvidia-cuda-toolkit, but I did not find any information on installing version 10.2 for ARM. Can you tell me how to install it?

I mainly develop software for an NVIDIA Jetson Xavier AGX (set up with JetPack 4.4).

For developing, I use a linux ubuntu 20.04 system running docker and nvidia-docker2.

I use QEMU to emulate the ARM-architecture as suggest by NVIDIA Container Runtime on Jetson · NVIDIA/nvidia-docker Wiki · GitHub. More information on Running and Building ARM Docker Containers on x86 | Stereolabs).

2

I found a similar question that already has been answered: Enabling Jetson Containers on an x86_64 Workstation using QEMU

To summarize, software for Jetson cannot be run on x_86 systems (even with docker and qemu) due to the hardware difference. There are ways to cross-compile (e.g. with the CUDA cross-compiler package installed by JetPack on the host), but the software can only be run on the Jetson.