JetPack 4.1.1 Developer Preview for Jetson AGX Xavier

NVIDIA has released JetPack 4.1.1 Developer Preview and L4T R31.1 for Jetson AGX Xavier Developer Kit, which is now available from NVIDIA DevZone. This update to the previous early access release includes:

  • TensorRT: new samples, DLA support for FP16 inference using LeNet for MNIST networks, and asynchronous DLA execution with IExecutionContext::enqueue()
  • cuDNN: now supports strided convolution with FFT tiling algorithms and improves performance for dilated and Winograd Transform convolutions
  • L4T BSP: stability improvements and IMX185 sensor now supported

Please refer to the JetPack Release Notes and L4T Release Notes for more information.

JetPack 4.1.1 Developer Preview components:

  • L4T R31.1
  • Ubuntu 18.04 LTS aarch64
  • CUDA 10.0.117
  • cuDNN 7.3.1.20
  • TensorRT 5.0.3.2
  • VisionWorks 1.6
  • OpenCV 3.3.1

Download JetPack…https://developer.nvidia.com/embedded/jetpack
Release Notes…https://developer.nvidia.com/embedded/jetpack-notes
L4T Release Notes…https://developer.nvidia.com/embedded/dlc/l4t-release-notes-31-1-0
L4T R31.1 Page…https://developer.nvidia.com/embedded/linux-tegra

Hi Dustin,

Does CBoot (included in JetPack 4.1.1 DP) support booting different kernels & DTBs from /boot directory on the eMMC rootfs via boot menu? If not, is there any projected date it’s planned to be implemented?

-albertr

Hi albertr, please refer to this post:

https://devtalk.nvidia.com/default/topic/1042509/jetson-agx-xavier/custom-u-boot-on-xavier-solved-/post/5298124/#5298124

Got it, thanks!

-albertr

Can I get any help to install caffe on Xavier ?

Hi pharatekomal, can you post your question as a new topic? Thanks!

EDIT: new topic here - https://devtalk.nvidia.com/default/topic/1044473/caffe-installation-on-xavier/#5299157

I’m running KDE Neon (Based on Ubuntu 18.04LTS)

jmc@jmc-x1c6:~$ uname -a
Linux jmc-x1c6 4.15.0-42-generic #45-Ubuntu SMP Thu Nov 15 19:32:57 UTC 2018 x86_64 GNU/Linux

If I run ./JetPack-L4T-4.1.1-linux-x64_b57.run

Creating directory _installer
Verifying archive integrity… All good.
Uncompressing JetPack 100%

I get a dialog stating I need to run Ubuntu, and that the installer cannot continue.
Does the run script have an option to ignore its (faulty?) OS check?

And if not, what do you use to check this, maybe I can fool it?..

Regards,
Jens.

Hi jensbv134, I looked through the .run file in a source editor, but was unable to find where the check was performed. Regardless, JetPack GUI was not designed or tested with KDE and may run into further issues there. The host PC system requires for JetPack 4 are vanilla Ubuntu 16.04 or 18.04, so it is recommended to use one of those standard distributions.

Although that is rather disappointing, thank you all the same for looking into it.

I’ll dig in and see if I can’t hack it, I’d rather try and deal with “further issues” than re-install my laptop and run gnome; That’s simply to much of an ask.

Note that you can install the L4T R31.1 BSP directly, without using JetPack.

If you were to run JetPack from a virtual machine to download the other packages (like CUDA, cuDNN, TensorRT, ect), you could copy those to your Jetson and install them manually with dpkg. JetPack downloads those packages to the jetpack_downloads/ folder

Note that some folks have limited success running the flashing & install operations from within a VM, but are typically able to download the packages at least. Normally it’s the flashing stage that fails from within a VM, and flashing JetPack-L4T from within a VM isn’t officially supported. However you could do the flashing from your native Linux machine and try installing the packages from JetPack in the VM. You can de-select the flashing step from the JetPack component manager GUI.

Error: libopencv-python libopencv-samples  cannot be installed on host.

My PC has opencv installation done.
While installating Jetpack, its giving issues to install opencv.
After searching for it on devtalk, came to know it may occur because of opencv installed on my PC.
Why such dependency? Any solution to solve it?