I want to analyze my mixed c++ cuda code. How to use NVIDIA Nsight Systems with python and pytorch?
The same way you use it for c++ but by changing the executable to be
python your_test_script.py IIRC
At least with nvvp that always worked for me…
What is your concern that you say it can not work?
My nsight systems can’t work. Perhaps the command line and working directory are wrong. What i wrote in that two settings is right?
When I used these settings for nvvp, it always worked for me.
Can you use nsight properly with other cuda programs?
Unfortunately, though my code can run successfully for nvvp, my server always crashes after running the code. I have to restart my server times by times. And some functions of nvvp can’t support my server, whose compute capability is bigger than 7.2. Thus, i want to use nsight system as a substitute. For nsight systems, it shows that “found no nvidia driver”. However, the same code can run successfully for pycharm and windows’ cmd. I have been confused for a long time.
How did you install Nsight Systems and is the CLI available via
nsys or Nsight Compute via