Multi-tasking performance GPU/CPU

Autonomous Machines Jetson & Embedded Systems Jetson AGX Orin
1

Hello,

I’d like to know if the Jetson AGX/NX Orin are capable of being a ‘standalone’ computing unit on a mobile system. More specifically, can I assume that Google’s Cartographer (3d Mapping) and modern neural network architectures (inference) for semantic segmentation etc. will run in parallel fluently or is it not recommended to run multiple tasks, including CPU heavy, on a single embedded AGx/NX orin unit? Right now we do have a quite old desktop PC on our Clearpath Ridgeback platform installed using a consumer GPU (titan) but we’re running into issues of drivers, the power consumption should be lower etc.

I also had a look for regular PCI express compatible gpus (not really embedded…) but it seems the current ones offered are all RTX cards which feels a bit over the top as we dont run rendering applications on the robots machine.

2

You can find some interesting project sharing at Latest Jetson & Embedded Systems/Jetson Projects topics - NVIDIA Developer Forums to gain some ideas if they’re capable, also refer to their Technical Specifications from Jetson Modules | NVIDIA Developer to help you making decision.

3

I suggest you check out this work: https://dl.acm.org/doi/abs/10.1145/3466752.3480101

The slowdown is measured on the CPU, GPU, and DLA. You can send a message to me if you have more specific questions.

1 Like
4

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.