Multiple Jetson AGX Orin & ZED Capture Card Duo

Hardware
,
1

Hi there.

I am currently setting up the hardware architecture for a project and looking for some insights from the community.

Context
Currently, we are testing with

  • 1x NVIDIA Jetson AGX Orin (64GB)
  • 1x ZED Link Capture Card Quad.

Goal
Understand the possibilities of connecting two Jetson AGX Orin Dev Kits via Ethernet.

  • Per Jetson: 1x ZED Link Capture Card Duo + 2x ZED X cameras (GMSL2).

Here are my questions:

  1. Latency: Is low-latency performance feasible between the two Jetsons via Ethernet for this setup?
  2. Sync: Since the Quad card supports internal hardware sync/triggering, what are the bottlenecks when trying to synchronize two separate ZED Link Capture Card Duos across two devices?
  3. Future Tech: Is there a status on NVIDIA Jetson Thor + ZED X applications? If that powerhouse is around the corner, we might skip the multi-Jetson headache.

Thanks for the help!

2

Hi @onit

This depends on how you make the device communicate.

The ZED Link Duo does not support hardware triggering, so hardware synchronization is not available with it.

You can contact ConnectTech or Advantech. They have Thor-ready solutions with our GMSL2 cameras.

3

Hi @Myzhar.

Thanks for your replies.
We have a few more questions, as Advantech seems to be a good direction.

Question 1
Are there any unknowns around the Stereolabs GMSL driver and its mapping to the Advantech MIB-742-AT device tree (especially the BSP)?

Question 2
The Advantech MIC-FG-8G capture card uses the MAX96712 deserializer. Even though this is not AGX Orin or Thor related, are there any similar issues like the one documented here: https://www.reddit.com/r/JetsonNano/comments/1q8iigk/zed_x_jetson_orin_nx_gmsl_driver_carrier_board/ ?

Question 3
Do you have experience with more JetPack 7 instabilities?

Question 3.1
We can see that there are nvidia-container-toolkit instabilities ( Jetpack 7.1 apt install issue nvidia-container - #15 by AastaLLL - Jetson Thor - NVIDIA Developer Forums ), though NVIDIA seems to resolve it quickly.

Question 3.2
Another one are continuous NVMe boot loops after flashing the OS: See Thor failing to boot - Jetson Thor - NVIDIA Developer Forums . What is your take on this?

Question 4
If there are any external NVIDIA SDKs involvements, how will we handle reported API friction and build failures (e.g., CCCL/RMM header incompatibilities) currently seen on the Blackwell toolchain? (see Build failure on Jetson Thor (JetPack 7.1) with Holoscan SDK 3.5.1 - Holoscan SDK - NVIDIA Developer Forums )

Question 5
Is the ZED SDK troubled by any data format migrations towards the Thor infrastructure (e.g. FP4 vs. FP8 data format handling)?

Happy to get your view on these points.