Any pre-requisits for running NSIGHT Compute?

Development Tools Nsight Compute
#1

I have installed NSIGHT Compute (NC) on a stand-alone test machine (not connected to network).

  • I am able to bring up the GUI.
  • I setup a project and have localhost at the target.

When I run, the application opens but the NC is unable to connect to the process.

  • All I see is “Trying to connect to process on host: localhost”.

I have NC on my dev machine and it is able to launch and connect to the application.

The application itself runs fine on the test machine.

I read the documentation (users manual) and there I did not see a “setup” or "requirements’ section outlining the need for other tools, drivers, SDKs, toolsets.

#2

Could you please let us know what operating systems you use for both your test and dev machines? It does help to be as specific as possible. When using the UI to start your application, do you use the “Interactive Profiling” activity, and do you launch the application via this instance of Nsight Compute (rather than e.g. attaching to a running application)?

Please also try the Nsight Compute command line interface (CLI) and see if that works any better for you.

Nsight Compute uses network ports on your system to communicate between the host (UI) and the target (app) process. It uses ports 4500-4510. If all ports in this range are used by other processes already, or Nsight Compute is not permitted to use them otherwise, this might result in a connection problem.

#3

Dev Machine:

  • Dell Laptop EliteBook
  • Quadro K1000M
  • Win7 Enterprise (Ver 6.1, Build 7601, SP1)
  • VS2015
  • Cuda10
  • CUDA Toolkit
  • NSIGHT VSE 6.0
  • NSIGHT Compute 1.0

Test Machine:

  • HP Z640
  • Win10 (Ver1809, Build 17763.107)
  • VS2015
  • Two GeForce RTX 2080 + NVLINK Bridge (SLI not enabled)
  • Latest GeForce Display Driver
  • NSIGHT Compute 1.0
  • NSIGHT Systems 2018.3.4.3
  • NSIGHT VSE 5.3

Just in case, I added specific rule to Windows Defender to allow communication via these ports (4500-4510). I still have the same issue.

Thank you for the support.

#4

I have the same problem, log keep spamming “Trying to connect to process on host: localhost”

Ubuntu 16.04
CUDA 10.1
Nsight Compute 2019.3
RTX2070

#5

Once more, when using the UI to start your application, do you use the “Interactive Profiling” activity, and do you launch the application via this instance of Nsight Compute (rather than e.g. attaching to a running application)?

Please try the Nsight Compute command line interface (CLI) and see if that works any better for you. Furthermore, check if the machine is able to resolve “localhost” to a valid IPv4 address.

Finally, note that it will only connect once the application actually uses the CUDA API.