Ncu does not launch process when called remotely, but does when called locally

Hi. I think I’ve run into a bug here.

When I try to use the nsight compute 2020.1 gui to launch ncu on a remote box, the ncu command itself launches (confirmed by observing process activity on the target), but the process to be debugged does not (likewise confirmed). On the local machine, the nsight log hangs at “Profiling”, with no additional information given. However, when I copy the command from the nsight gui and run it directly in an ssh session on remote, both ncu and the target process run as expected.

info:
local software: nsight compute 2020.1
local os: Fedora 32
target os: Ubuntu 18.04
target gpu: Titan RTX
target cuda: 10.2

I can provide more info as required.
Thanks.

From your description, I think it is supposed to work, I can’t see any obvious mistakes, especially since you verified that it works on the remote machine with the deployed version of the tool. The only thing you could double-check are the environment settings, i.e. if you have set/unset specific env vars after the ssh login that you don’t set when remote launching, or that you use a different working directory?

If those hints don’t help, you can file a bug via https://developer.nvidia.com/nvidia_bug/add
Please use the Area: DesignWorks, category: Developer Tools.

Thank you.