Advantec MIC730-AI

[flavio.mello]

Hi there,

I received a MIC730-AI (MIC-730AI - AI Inference System based on NVIDIA® Jetson AGX Xavier™ - Advantech) with an NVIDIA Jetson AGX Xavier inside.

It comes from factory with no CUDA installed, no cuDNN. It has JetPack 4.3.

Using this configuration, is there any way to get benefits from the GPU? As far as I know, it is useless. It is just an ARM processor with no use of GPU. Am I wrong?

Thanks in advance,

Flávio Mello

[andres.artavia]

Hello @flavio.mello ,
Is there a reason you cannot install software on the device?
If not, you have a couple of options, use the sdk manager to install cuda and other sdk components or use apt. For both you can do it via ssh.
The easier option with gui is sdkmanager.
You need a pc with the sdk manager installed. Preferably Ubuntu 18.X or 16.X otherwise you cannot use the normal sdkmanager to install JP 4.X versions, and you will need to use docker, like [here].(Docker Images :: NVIDIA SDK Manager Documentation)
On this screen select the target JP version and the target hardware. In your case JP 4.3, you may need to open sdkmanager in console with “sdkmanager --archivedversions” if it doesn’t show up.

Then since you need only the sdk components unselect the Jetson-linux option, and select the Jetson sdk components, as shown in the image

Then it will promt for a connection, then select ethernet and fill up the fields.

Then it should install the packages.
If you want to use apt, this guide, can help but it has more steps.

Regards,
Andres
Embedded SW Engineer at RidgeRun
Contact us: support@ridgerun.com
Developers wiki: https://developer.ridgerun.c om/
Website: www.ridgerun.com

[flavio.mello]

So, I was trying an alternative do SDK Manager because I had to have access to the mainboard. But I gave up and break the warranty seal of the Advantec box in order to access the USB maintainance slot and REC/RST buttons. Now I installed everything through SDK Manager.

The OpenGL 4.1.1 that is installed by SDK Manager will use GPU or it needs to be recompliled with CUDA access?

[andres.artavia]

@flavio.mello
It should have it, I do know that OpenCV is not included with Cuda by default, but OpenGL should. Have a look at this example and see it it works. Or if you want somehting a bit more complex.
Also refering to opening the carrier, for the install using the sdk manager at least for the sdk components, that could be done via ssh, you don’t need to go into the recovery port. But if you want to completely reinstall jetpack, then it’s needed.

Regards,
Andres
Embedded SW Engineer at RidgeRun