Just upgraded to the latest and this is really slow

Robotics & Edge Computing Jetson & Embedded Systems Jetson TX2
1

I am trying to run a few TensortRT samples here like like
./detectnet-camera pednet
and
./segnet-camera
and noticed the upfront build time was really long and
after segnet-camera started it was VERY slow. like .5 fps

I am not sure why its always pausing for a long time at building CUDA engine

Can anyone shine a light on that…

2

Hi,

Please remember to maximize the CPU/GPU clock first.

sudo ./jetson_clocks.sh

This observation is under our expectation.

When launching a new model, TensorRT takes some time to optimize the implementation for inferencing.
Once having the optimized PLAN, TensorRT can de-serialize the PLAN without re-compiling.
Then it will be much faster the first launch.

Thanks.