Could L4T-cuda r35 or r32 run in Jetpack 6.0 docker?

Robotics & Edge Computing Jetson & Embedded Systems Jetson Orin Nano
,
1

I want to run L4T containers with different versions CUDA (12.2/11.4/10.2) in Jetpack 6.0 docker. Could jetpack 6.0 be compatible backwardly ?
In this thread ( How to enable CUDA on docker container under JetPack 6.0? ), I saw it requests the same version btw jetpack and L4T. But I still want to know is there another way to support multi versions of CUDA in jetpack 6.0 docker ?

Thanks for your supports!

3

Hi,

It is possible to install a higher CUDA library on JetPack 6.0.
So you can use12.2+ CUDA library.

But backward compatibility is not supported.
Thanks.

4

Thanks for your help!

I still have some questions.

  1. Jetpack 6.0 supports CUDA 12.2+ in docker containers. And does it mean L4T with CUDA 12.5 is supported in Jetpack 6.0 docker container?
  2. How about forward compatibility? e.g. L4T with CUDA 12.2 in Jetpack 5.1.2 docker container with CUDA 11.4

So, could I make a conclusion that for GPU passthrough in jetpack docker container, the major version of CUDA must be the same between Jetpack and L4T-cuda? e.g. 12.2 and 12.5
Or, the major and minor version must be the same. e.g. 12.2 and 12.2.1

5

Hi,

1. Yes, CUDA 12.5 can work on JetPack 6 inside or outside the container.

2. This is more related to the GPU driver.
JetPack 5 container cannot work on JetPack 6.

So if you want to have CUDA 11 and CUDA 12 container at the same device.
Please start with the same base image (ex. r35 or r36) and upgrade the CUDA to 12+.
Running a r36 container on r35 BSP is not supported.

Thanks.

6

Thanks for your professional reply!

Could you give more detail or the upgrading guide for CUDA 11 and 12 container at the same device?

7

Hi,

Sorry, there is a missing info.
JetPack 5 and JetPack 6 use different OS systems (JetPack 5: 20.04, JetPack 6: 22.04), so this yields an extra limit for the CUDA library.

JetPack 5: CUDA 11.4 - 12.2
JetPack 6: CUDA 12.4+

So if you want both CUDA 11 and 12, please use JetPack 5 to set up your Orin (r35).

Then you can get the CUDA 11 container in the below link:
https://catalog.ngc.nvidia.com/orgs/nvidia/containers/l4t-base/tags
nvcr.io/nvidia/l4t-base:35.3.1

Please use the above image as the base and build a custom CUDA 12 container with the below command:

Thanks.

9

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