JetPack 3.1 — L4T R28.1 released for Jetson TX1/TX2

Autonomous Machines Jetson & Embedded Systems Jetson TX2
1

External Media

Hello everyone, we’re pleased to announce that JetPack 3.1 with L4T R28.1 has been released.

JetPack 3.1 is a production software release for both Jetson TX1 and TX2, with Linux kernel 4.4 shared between both devices. In addition to bug fixes and improvements in the camera stack, JetPack 3.1 also includes updates to TensorRT 2.1 and cuDNN 6.0, delivering twice the inferencing performance for latency-sensitive applications. See our Parallel ForAll article, JetPack 3.1 Doubles Jetson’s Low-Latency Inference Performance.

View the full JetPack 3.1 Release Notes here, and the Release Notes for L4T here.

Package Contents

  • L4T BSP 28.1 64-bit with Linux kernel 4.4
  • Ubuntu 16.04 LTS aarch64 sample root filesystem
  • CUDA 8.0.84
  • cuDNN v6.0
  • TensorRT 2.1
  • OpenCV 2.4.13
  • VisionWorks 1.6

Release Notes…https://developer.nvidia.com/embedded/jetpack-notes
Download JetPack…https://developer.nvidia.com/embedded/jetpack
Software Overview…https://developer.nvidia.com/embedded/develop/software
28.1 Release Page…https://developer.nvidia.com/embedded/linux-tegra
Archive Page…https://developer.nvidia.com/embedded/linux-tegra-archive
Embedded DLC…https://developer.nvidia.com/embedded/downloads

2

hoorah!

3

Good !

4

cuDNN 6 is great. 2x improvement in start-up/batch-1 latency is great.
And user plug-in layers is super great! (#YOLO support!)

5

Umm, I have to try it on Tegra X2 to see how great it is.
Have you tried YOLO on Tegra X2 btw? If so, would you please let me know your achieved time performance in FPS(video frames per second).
I tried Tegra X1 once on YOLO and its time performance was as much as 12 FPS , if I am not mistaken.
My current platform is tegra X2 so I am curious to know.

Thanks for the reply btw ;)

6

Appreciate your hard work nvidia. Have been waiting for this. Hopefully we can get multiple cameras working concurrently.
Thanks

7

I wonder whether the “jetson-inference” code also works with the newly released TensorRT 2.1 on TX1/TX2.

[url]https://github.com/dusty-nv/jetson-inference[/url]

8

Hello,
I just installed the CUDA, cudnn and TensorRT from the new Jetpack and couldn’t compile detectnet inference on my Jetson TX2.
It appears that some API was changed (I get a ‘Dims3 in namespace nvinfer1 does not name a type’ error). I would like to know how to bypass this.
Besides that, when I tried to reinstall the previous Jetpack version (3.0) to get my detectnet working again, the installation failed at the stage of populating the component manager (got an error without any more information).
Thanks,
Haggai

9

I will be checking in the updated jetson-inference to reflect the changes shortly. Currently I am wrestling if I can make master backwards-compatible (i.e. using #ifdefs) so folks still on JetPack 3.0 / JetPack 2.3 can still compile.

10

Hi dusty_nv,

Sorry for reposting. This being a sticky thread, I thought the response will be quicker.

The kernel source available at [url]https://developer.nvidia.com/embedded/dlc/l4t-sources-28-1[/url] is incomplete. Missing TX2 related files. Can you provide the correct source please.

11

bugreport: twice it has happened that on power on neither monitor nor network seems working.
Got resolved first time with reflash. Second time with multiple power off - power on attempts.
What could be the cause?

12

I got the same error with my fresh TX2. I don’t like this problem… I was already struggled a lot of problem today, and when i was prefer download the 3.1 update i thought this is the much more stable one… but no .nvidia still needs to be work on it. I hope your way is working on me. Thanks for sharing.

Edit: Yeah it worked for me. I just unplug the all cables and just re-plug the ethernet and keep resetting until it works. Thanks man!

13

I was able to flash Jetpack3.1 successfully on TX2. After that I was creating a bootable SD Card for TX2.

For jetpack3.0 I followed this(Run Jetson TX1 from SD Card - YouTube) link and was able to boot TX2 from SD Card rather than EMMC. But with JetPack3.1 it is not booting from SD Card after following the same procedure. It boots for 13 seconds, then says ‘Rebooting in 5 seconds’ and this cycle continues again and again.

It is booting successfully from EMMC.

Earlier in Jetpack3.0 extlinux.conf file used to look different. Can this be an issue here?

Older Version:
TIMEOUT 30
DEFAULT sdcard

MENU TITLE p2771-0000 eMMC boot options

LABEL sdcard
MENU LABEL SD Card
LINUX /boot/Image
FDT /boot/dtb/tegra186-quill-p3310-1000-c03-00-base.dtb
APPEND fbcon=map:0 net.ifnames=0 console=tty0 OS=l4t console=ttyS0,115200n8 memtype=0 video=tegrafb no_console_suspend=1 earlycon=uart8250,mmio32,0x03100000 gpt tegraid=18.1.2.0.0 tegra_keep_boot_clocks maxcpus=6 android.kerneltype=normal androidboot.serialno=0335115020673 vpr_resize root=/dev/mmcblk1p1 rw rootwait

LABEL primary
MENU LABEL Internal EMMC
LINUX /boot/Image
APPEND fbcon=map:0 net.ifnames=0 console=tty0 OS=l4t console=ttyS0,115200n8 memtype=0 video=tegrafb no_console_suspend=1 earlycon=uart8250,mmio32,0x03100000 gpt tegraid=18.1.2.0.0 tegra_keep_boot_clocks maxcpus=6 android.kerneltype=normal androidboot.serialno=0335115020673 vpr_resize root=/dev/mmcblk0p1 rw rootwait

Current Version:
TIMEOUT 30
DEFAULT sdcard

MENU TITLE p2771-0000 eMMC boot options

LABEL sdcard
MENU LABEL SD Card
LINUX /boot/Image
APPEND ${cbootargs} root=/dev/mmcblk1p1 rw rootwait rootfstype=ext4

LABEL primary
MENU LABEL EMMC
LINUX /boot/Image
APPEND ${cbootargs} root=/dev/mmcblk0p1 rw rootwait rootfstype=ext4

if I try the older version of extlinux.conf it gets stuck while rebooting. Please Help

14

If you haven’t already, please create a separate topic about it for reporting bugs.

15

@dustin feel free to delete from that thread the mentioning of the issue - I have created a separate thread for the issue.
Thanks

16

Hi guys, support for TensorRT2 is working now in jetson-inference master with commit e40bd6

For backwards compatibility, it’s still building for previous JetPack’s still on TensorRT1 using some macros I added.
There were no outward-facing changes to the vision primitive APIs themselves.

17

Has anyone successfully rebuilt the R28.1 kernel yet? I’ve had issues and so has a fellow developer but I have not seen any issues reported yet…

If any one has did they need to modify any of the steps in the documentation. I’m following these very closely with the provided compiler and getting make sub failed error 2…

18

Hi ross,

Check this link : [url]https://devtalk.nvidia.com/default/topic/1019687/jetson-tx2/jetpack-3-1-kernel-source-tag-problem/post/5193793/#5193793[/url]

There were some errors in the kernel source earlier. But it appears to have been fixed now. You could try downloading the source again. This time you shouldn’t get any build errors.

19

Just a heads-up…the issue of apt-get update overwriting libglx.so is back. Anyone wanting to fix this can copy (from the driver package) “nv_tegra/nvidia_drivers.tbz2” over to “/” of the Jetson, and upack it to fix this (verify good/bad with “sha1sum -c /etc/nv_tegra_release”):

# On the Jetson, after copy of nvidia_drivers.tbz2 to "/"...
cd /
sudo tar xvfj nvidia_drivers.tbz2
# Verify fixed:
sha1sum -c /etc/nv_tegra_release
20

Whoa, seriously? ;-(