I have had a regression where new I a system built last year which runs without sheering and with acceptable latency sb3d new installs have 300ms of latency or more and sheering.
my application is an egl app based off of the argus_camera app in the jetson mmapi examples.
my hardware is:
tx2nx
capable robots baseboard
2 vision components Imx412c
compiz consumes 140% cpu or more and xorg consumes 55% cpu or more.
Test case 2:
reboot
open terminal
run top
run argus_camera
Again, compiz consumes > 150% cpu Xorg consumes > 50% cpu.
Latency is >300ms. (measured by comparing times on a picture which contains a stopwatch and the image of the stopwatch)
Both systems are built with 32.7.2.
On a separate note:
In building test cases I tried to use sdkmanager to setup my system so that I had a starting point that we can all duplicate. Unfortunately I am unable to use sdkmanager because if I select jetson sdk components The the OS is downloaded and then when it attempts to download the SDK components it fills the flash on the target. Once the flash is filled on the target the target system becomes unstable and needs to be flashed again.
The only sdk component that I want is the MMapi. In the past I remembre being to select SDK components individually. Can I do that now?
after reboot tun the same video and note cpu load in compiz goes up drastically. Moving windows in chrome also gets jerky and the system performance in general degrades.
the testsetup for this test is a TX2NX on a carrier board built by Capable Robot https://capablerobot.com/products/nx-baseboard/
No cameras are used in the test and no kernal or device tree changes have been made .
You should be able to duplicate this easily.
also can you give me a link to earlier versions of the mmapi?
can you give me instructions for cloning a working system (All partitions) and loading it on another target.
Hi,
The previous comment is updated to add sudo jetson_clocks. Please give it a try. It is not easy to check CPU usage if clocks are varying. Please fix it to maximum value and check tegrastats. You may think loading is high but actually CPU clock is at low value.
You or your engineers should be able to duplicate this problem this problem from the information in the attached file. there are less than 5 steps to duplicate the issue.
Please answer the following questions:
how do I clone all partitions from my working system and reload them on a new board?
where can I download earlier versions of nvidia-l4t-jetson-multimedia-api from?
Hi,
Do you observe the issue on Xavier developer kit? Seems like the version is mismatched so it does not work properly. Are you able to install SDK components through SDKManager and check if the issue is still present?
Dane,
Have you or your engineers tried to duplicate this?
As you will note in the first post in this thread, I cannot use SDKmanager because after an hour it has a fault when it fills the eeprom. There is no way to select which components to install. it is all or nothing… and all is too big.
I do not have a Xavier development kit. The carrier board that I am using is does not require any device driver or device tree changes unless you want to access the mipi camera via i2c camera.
The test plan I have given specifically does not use the camera and uses the 32.7.3 right out of the box. The only command that I give after loading 32.7.3 is to load the MMAPI using apt-get. I do not know what gets loaded with this command. Please try it and let me know if your results match mine.
The files that I have sent you should give you enough information that you should be able to duplicate the problem and figure out what incompatible versions I am getting. The only thing that I do to get the problem is to load the MMAPI using apt-get. I do not have the knowledge or tools to debug this. I do not know what version of the MMAPI is being brought in and I do not know what dependencies are being loaded.
I have a product which is supposed to ship. It cannot ship until the performance of the OS has been restored.
Hi,
We don’t observe the issue on Xavier developer kit, so suspect you don’t install correct version of the packages since there are multiple r32.7.x releases.
You should be able to generate a nvidia-jetpack script for your custom board. It would be great if you can refer to the post and give it a try. The post is the script for r32.7.1
We will also check and try to make a script for r32.7.3.
Again, this problem is with tx2nx not Xavier
I do not have a Xavier development kit.
Did you try to reproduce this on a tx2nx or a Xavier? I suggest that you try with a tx2nx.
As I have clearly stated I am using 32.7.3
All of the exact commands that I used to duplicate the problem for you are listed in the files that I have downloaded. Yes it seems that something incompatible with tx2nx is being loaded. I am asking for your help to figure out what it is.