I am developing an python based face recognition model training on NVIDIA AGX Orin Developer Kit. I intend to observe the utilization of the the CPU memory, GPU memory, SM, CUDA cores, CPU cores with Pytorch framework, Pytorch-CUDA framework and TensorRT libraries.
I need the support for python based profiling for model training. Kindly guide/ share the tutorial links for python based profiling for model training.
For information about support for python profiling in Nsight Systems see User Guide — nsight-systems 2025.1 documentation
(that link is directly to the python programming section, the text is just munged by the forum software).
Thank you.
I just want the help with Nsight Compute using Linux Target Platform as marked with pink colour in the Screenshot given below:
I want to profile the GPU memory utilization when working with python code utilizing 1. Pytorch-CUDA 2. TensorRT.
Can a Python code utilizing CUDA be profiled on NSIGHT Compute and How?
Moving your request to that module, and pinging @mstrengert to comment or assign.
Hi, @snehal2
Python profiling is supported in Nsight Compute also. You need to specify “python XXX.py” as target application.
Thank you. I will try the same.