TensorRT 8 to 10 migration (orin nx) c++ is a nightmare :-(

[dannykario]

I am trying to migrate a working c++ code from trt7/8 to trt10 (on orin nx).

It seems most of the backword compatibility is broken (few examples - IFullyConnectedLayer, Enqueue, addConvolutionNd in run time and many more) , and I find very little documentation for usage or migration. Are there any resources/examples/guide for this migration ?

As an alternative - can I run a docker in the orin with trt8 without hurting performance ?

Thanks for the help ! Best
Danny

[AastaLLL]

Hi,

Please find the comment in original topic:

JetPack 6.1, downgrade TRT from 10 to 8 (jetson orin nx)

Hi, Working on jetson orin nx. Installed the JP6.1, the trt10 broke all my c++ TRT code, with no backward compatibility. Is there a way to run a docker with TRT8 on the JP6.1 , where the container is accessible to the GPU and (hopefully) to the DLA ? Or Install TRT8 on the jp6.1 in parallel/instead of the trt10 ? Thanks for the help !

Thanks.