Remote debugging

Developer Tools Nsight Visual Studio Code Edition
,
1

Topic: Help with Remote Debugging on Jetson Orin Nano using Nsight VS Code

I’m having trouble setting up remote debugging for the Jetson Orin Nano using Nsight Visual Studio Code.

We were able to build the application with debug symbols and successfully debug it using Nsight VS Code directly on the Jetson Orin Nano (i.e., when running Nsight locally on the device).

However, I want to set up remote debugging from a host PC running Ubuntu. In this setup:

  • The source code resides on the host PC.
  • The compiled binary with debug symbols is on the Jetson Orin Nano.
  • I want to use Nsight VS Code on the host PC to connect via SSH and perform debugging on the Jetson.

When I attempt this configuration, it doesn’t work as expected.

Is this type of setup officially supported?
If so, how can I properly configure Nsight VS Code to enable remote debugging where the source code is on the host and the binary executes on the Jetson Orin Nano?

Also, is it possible to perform GPU kernel debugging and collect full GPU profiling reports (e.g., using Nsight Compute or Nsight Systems) on the Jetson Orin Nano remotely from the host PC?

Any guidance, best practices, or setup instructions would be greatly appreciated.
Thank you!

2

I have the same problem. Please let me know if you find a solution. Thank you.

3

Based on your situation, I recommend to use CUDA GDB Server: Launch.
To use that you need to follow below steps: (Take cuda sample matrixMul as example)
you can scp/cp binary to your target machine

  1. for launch.json, add Configuration: such like
{
    "name": "CUDA GDB Server: Launch",
    "type": "cuda-gdbserver", // for QNX "cuda-qnx-gdbserver"
    "request": "launch",
    "server": "cuda-gdbserver",
    "program": "/home/swqa/NVIDIA_CUDA-12.9_Samples/Samples/0_Introduction/matrixMul/matrixMul", // Based on your source code and binary
    "target": {
        "host": "127.0.0.1", // host ip
        "port": "12345" // port which cuda-gdbserver start on step2)
    },
    "sysroot": "", // set "target://" to save time for transfer libraries
    "debuggerPath": "/usr/local/cuda/bin/cuda-gdb", // Sepcify cuda-gdb path on this Rubicon Launched Machine "/usr/local/cuda/bin/cuda-gdb"
    "breakOnLaunch": true
},

for more information please refer to Getting Started with the CUDA Debugger :: NVIDIA Nsight VSCE Documentation

  1. ssh to your Jetson Orin Nano target, use cuda-gdbserver to launch the binary
cuda-gdbserver :12345 ./matrixMul
  1. On your host PC (which Nsight VSCode Running on, launch.json on ) start debugging with CUDA GDB Server: Launch