How can I estimate the overall execution time of CPU and GPU separately?

I am using Nsight system/Nsight Compute 2021.2.4 to profile my pytorch GCNs applications in x86 Linux with CUDA 11.4. I want to estimate the application execution time on GPU and CPU separately, but the Nsight system profiling report is seemly not provide the execution time comparsion between CPU and GPU.

How can I achieve this goal and what should I do ? Thanks.

1 Like