Error using nsight system to analyze pycuda

Using the nsight system CLI to analyze the pycuda program in Windows, an error occurs as shown in the following figure. And the analysis is incomplete. Does anyone know how to fix it?


It looks to me based on the initial screen shot, that you hit issues in your python runtime.

Does the program run without Nsight Systems?

This program can run without the nsight system, and the nsight computer can record it as normal. I use CUDA11.7 and nsight system2023.2.1

@rknight can you take a look at this?

Hi jychen2021,
Can you try the following nsys CLI commands to see if the issue continues to happen?

nsys profile --trace=none --sample=none --cpuctxsw=none C:\Users\Jinyun.conda\envs\jinyunwork\python.exe mm.py

nsys profile --trace=cuda --sample=none --cpuctxsw=none C:\Users\Jinyun.conda\envs\jinyunwork\python.exe mm.py

Hello, I tried the nsys CLI commands you suggested, but I got the following error result

(Jinyunwork) PS D:\mypython_workstation\separate_test_file\mutil_stream> nsys profile --trace=none --sample=none --cpuctxsw=none C:\Users\Jinyun.conda\envs\jinyunwork\python.exe mm.py
Dynamic exception type: class std::range_error
std::exception::what: bad conversion
(Jinyunwork) PS D:\mypython_workstation\separate_test_file\mutil_stream> nsys profile --trace=cuda --sample=none --cpuctxsw=none C:\Users\Jinyun.conda\envs\jinyunwork\python.exe mm.py
Dynamic exception type: class std::range_error
std::exception::what: bad conversion

I have used administrator rights

Can you try the following nsys CLI command without admin rights?

nsys profile --trace=none --sample=none --cpuctxsw=none C:\Users\Jinyun.conda\envs\jinyunwork\python.exe mm.py

I got the same result

(Jinyunwork) PS D:\mypython_workstation\separate_test_file\mutil_stream> nsys profile --trace=none --sample=none --cpuctxsw=none C:\Users\Jinyun.conda\envs\jinyunwork\python.exe mm.py
Dynamic exception type: class std::range_error
std::exception::what: bad conversion

There is clearly a bug in nsys. Can you collect a log and share it with us to help us debug this issue?

In the Nsight Systems target-windows-x64 directory, there is a file named nvlog.config.template. Rename this file nvlog.config and then run the

nsys profile --trace=none --sample=none --cpuctxsw=none C:\Users\Jinyun.conda\envs\jinyunwork\python.exe mm.py

command again.

Afterwards, there should be a file named nsys-ui.log in your working directory. Please share that file (nsys-ui.log) with me.

Thank you for your help!

My nvlog.config.template file is located in host-windows-x64, and I changed the file name to get the nsys-ui.log file as well
nsys-ui.log (1.7 MB)

I apologize for the slow reply. I was out of the office last week.

Thank you for the log. But, we actually collected the wrong log which is my mistake for not providing more information.

Can you move (not copy) the nvlog.config file from the host-windows-x64 directory to the target-windows-x64 and rerun the

nsys profile --trace=none --sample=none --cpuctxsw=none C:\Users\Jinyun.conda\envs\jinyunwork\python.exe mm.py

command again? A new nsys-ui.log file should be generated in the working directory. Please share the new log file with me.

Thanks a ton for your help in debugging this issue!