We’re pleased to announce that JetPack 3.3 is now available for Jetson TX2/TX2i and Jetson TX1!
The highlight of this release is TensorRT 4.0, enabling support for TensorFlow’s TensorRT integration feature. Also, cuDNN has a small update to support the new TensorRT version, and all other JetPack components remain unchanged from JetPack 3.2.1. Please view the Release Notes.
Release notes are still only for JetPack3.2.1. Will there be separate release notes? I assume the L4T notes are the same, but I was thinking perhaps JetPack itself has new release notes.
Looks like it has the same CUDA 9.0 as previous releases… Is there a timeline for releasing a new CUDA libraries with fix for >4GB continues memory allocation issue on aarch64 platform?
The Python API for TensorRT isn’t supported on Jetson, so it isn’t included, however many Python-based models can be imported (via the UFF and new ONNX exporters), and TensorFlow-TensorRT integration is available (TF-TRT).
JetPack is just a front end. It’ll download whatever is current for that JetPack version for whichever device it is flashing.
That said, the earlier releases had different software between the TX1 and TX2. More recent releases (which JetPack3.3 will use) share the same sample rootfs. You’ll find JetPack creates a TX1 and TX2 directory during the flash of the TX1 or TX2, respectively.
linuxdev is correct, you only need to download the JetPack installer once regardless of if you are on Jetson TX1, TX2, or TX2i. JetPack will then allow you to select which device you are using, and will automatically download the appropriate packages for you to your host PC as linuxdev mentioned.
You may say if you have run these commands on host or on Jetson.
If it was on Jetson, you may upgrade, as long as you keep with Ubuntu16.
Also note that big upgrades sometimes make compiz cache wrong. So, if after rebooting your user cannot log in (GUI start fails or it start without some bars), then just login as the other user (if you’ve upgraded as user nvidia, then use user ubuntu) and delete compiz cache for user nvidia (or user ubuntu if you upgraded as user ubuntu):
You should always update. But…before you reboot, be sure this shows all ok:
sha1sum -c /etc/nv_tegra_release
There are actually two copies of libglx.so, and one can get overwritten (it is a rare update). If this happens simply copy the “ok” one to the bad one. Failing to do so before rebooting means the GUI won’t work until it is put back in place. The two specific lines of that command:
/usr/lib/xorg/modules/extensions/libglx.so: OK
/usr/lib/aarch64-linux-gnu/tegra/libglx.so: OK
If a package update got in the way, it will be the xorg module one which went bad. So: