Cannot launch Nsight Compute, connection error detected communicating with target application, GTX 3090

I am trying to launch a CUDA app. I could use Visual Profile on GTX 1050. Now I need to run it using GTX 3090 and use Nsight Compute. I see following errors:

I am attaching screen shots of configurations. I am trying to launch using Interactive profile and Profile in the connection setting.


The error messages indicate that the profiled target process (your app) terminated with a SIGABRT signal. Without knowing the application in detail, it’s not possible to say right away if this is an application problem, or caused by the attached tool. Some suggestions to check and try

  • Ensure that the application under the tool is executed in an environment that is similar to the one when running it by itself. If you typically run your app from a shell environment, make sure that e.g. the same environment variables are set in the Nsight Compute UI.

  • Try profiling using the ncu command line from a shell

  • Reduce the number and type of selected metrics. Different metrics are collected using different mechanisms. Try the following experiments on the CLI:

    ncu --metrics gpc__cycles_elapsed.sum <app> <app args>
    ncu <app> <app args>
    ncu --set full <app> <app args>

You may add -o my_report to each of these commands to create an ncu-rep report file that can later be loaded in the CLI or UI instead of printing to stdout.

Please also provide the NVIDIA driver version you have installed on this system.

Thanks @felix_dt ,

Here is my driver version:Driver Version: 470.86;

I found error that in the arguments I typed a file name which is not in the directory.
I fixed it by entering a file name (ending with cgf) into the working directory.


This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.