TensorRT on-device cross compilation compatibility?

Hi,

I’m trying to work out what Jetsons I need to buy to support a range of customers. I can’t find any concrete information about whether models compiled on one Jetson can work on another, or if cross compilation is compatible. Can someone please answer the following scenarios:

Can I use a model compiled on a Xavier NX 8GB on a Xavier NX 16GB?
Can I use a model compiled on a Xavier NX on a Xavier AGX?
Can I use a model compiled on a Orin AGX on a Xavier AGX?
Can I cross-compile a model on an x86 system for use on a particular Jetson platform?

What other caveats are there? I don’t want to buy every possible Jetson just to compile models correctly.

Kind regards,

Thom

Hi,

The serialized engines are not portable across platforms or TensorRT versions.

For your use case, you can try the emulation feature on Orin devkit.
It can emulate other Jetson Orin modules that might meet your requirement.

Thanks.

Thanks for the reply and the link. Just to confirm: there’s no compatibility even between the 8GB and 16GB Xavier NX? And other than the Orin devkit, there is no way to cross compile for other modules?

Hi,

Unfortunately no.
TensorRT engine is not portable so it’s required to be compiled on the exact target.
However, it’s possible to use devkit to emulate other modules.

Thanks.

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