Can a TensorRT engine built on a desktop GPU be copied directly to a Jetson?

An engine contains choices made for a particular runtime and hardware environment. Check the support matrix for the exact TensorRT versions and platform combination. Special compatibility modes have defined limits; they should not be assumed to make an arbitrary desktop plan portable to Jetson.

Keep an exportable source model and reproducible build configuration. Build or validate the optimized engine in a supported target environment, including any custom plugins and dynamic input profiles. Use the failure log to distinguish a version mismatch from a missing plugin or unsupported operation.

Run the same evaluation inputs after rebuilding and compare outputs with the reference model. Then measure target memory use and latency at the intended precision. Successful deserialization is a compatibility check, not proof that the model's numerical behavior or resource budget is acceptable.